Algebraic Models in Geometry - School of Mathematics

Oxford Graduate Texts in Mathematics Series Editors R. Cohen S.K. Donaldson S. Hildebrandt T.J. Lyons M.J. Taylor

OXFORD GRADUATE TEXTS IN MATHEMATICS Books in the series 1. Keith Hannabuss: An introduction to quantum theory 2. Reinhold Meise and Dietmar Vogt: Introduction to functional analysis 3. James G. Oxley: Matroid theory 4. N.J. Hitchin, G.B. Segal, and R.S. Ward: Integrable systems: twistors, loop groups, and Rieman surfaces 5. Wulf Rossmann: Lie groups: An introduction through linear groups 6. Qing Liu: Algebraic geometry and arithmetic curves 7. Martin R. Bridson and Simon M. Salamon (eds): Invitations to geometry and topology 8. Shmuel Kantorovitz: Introduction to modern analysis 9. Terry Lawson: Topology: A geometric approach 10. Meinolf Geck: An introduction to algebraic geometry and algebraic groups 11. Alastair Fletcher and Vladimir Markovic: Quasiconformal maps and Teichmüller theory 12. Dominic Joyce: Riemannian holonomy groups and calibrated geometry 13. Fernando Villegas: Experimental Number Theory 14. Péter Medvegyev: Stochastic Integration Theory 15. Martin Guest: From Quantum Cohomology to Integrable Systems 16. Alan Rendall: Partial Differential Equations in General Relativity 17. Yves Félix, John Oprea and Daniel Tanré: Algebraic Models in Geometry

Algebraic Models in Geometry Yves Félix Département de Mathématiques Université Catholique de Louvain-la-Neuve Louvain-la-Neuve Belgium [email protected]

John Oprea Department of Mathematics Cleveland State University Cleveland, USA [email protected]

Daniel Tanré Département de Mathématiques Université des Sciences et Technologies de Lille Lille, France [email protected]

1

3

Great Clarendon Street, Oxford OX2 6PD Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York © Y. Félix, J. Oprea, and D. Tanré 2008 The moral rights of the authors have been asserted Database right Oxford University Press (maker) First published 2008 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose the same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloging in Publication Data Data available Typeset by Newgen Imaging Systems (P) Ltd., Chennai, India Printed in Great Britain on acid-free paper by Biddles Ltd., King’s Lynn, Norfolk ISBN 978–0–19–920651–3 ISBN 978–0–19–920652–0 (Pbk) 1 3 5 7 9 10 8 6 4 2

To Agnès. To mon Papillon: Kathy. To the Magnificent Seven: Camille, Elsa, Amélie, Mathieu, Thibault, Lucie, Maude.

This page intentionally left blank

Preface Rational homotopy theory originated with the work of D. Quillen and D. Sullivan in the late 1960s. In particular, Sullivan defined tools and models for rational homotopy inspired by already existing geometrical objects. Moreover, he gave an explicit dictionary between his minimal models and spaces, and this facility of transition between algebra and topology has created many new topological and geometrical theorems in the last 30 years. An introduction to rational homotopy whose main applications were in algebraic topology was written some years ago. Because of recent developments, it is clear that now is the time for a global presentation of some of the more representative geometrical applications of minimal models. That is the theme of this book. Before giving an overview of its content, we present the basic philosophy behind the theory of minimal models. As Sullivan wrote in the introduction of Infinitesimal Computations in Topology: We have suggested here that one might therefore recall the older methods of differential forms, which are evidently quite powerful.

When de Rham proved that H ∗ (ADR (M)) ∼ = H ∗ (M; R) for the differential algebra of differential forms ADR (M) on a manifold M, it immediately provided a link between the analysis on and the topology of the manifold. Sullivan is suggesting in his remark that even within the world of topology, there is more topological information in ADR (M) (henceforth called the de Rham algebra of M) than simply the real cohomology. For a compact connected Lie group G, there exists a subdifferential algebra of bi-invariant forms, I (G), inside the de Rham algebra ADR (G), such that the canonical inclusion I (G) → ADR (G) induces an isomorphism in cohomology. This is the prototype of the process for models: namely, we look for a simplification MM of the de Rham algebra with an explicit differential morphism MM → ADR (M) inducing an isomorphism in cohomology, exactly as bi-invariant forms do in the case of a compact connected Lie group. In order to implement this strategy, we first have to make precise what a “simplification” means. In the de Rham algebra, we might suspect that some information is contained in two different entities: the product of

viii

Preface

forms, which tells us how two forms can be combined together to give a third one and the exterior derivative of a form. In a model, we kill the information coming from the product structure by considering free algebras ∧V (in the commutative graded sense) where V is an R-vector space. This pushes the corresponding information into the differential and into V where it is easier to detect. More precisely, we look for a cdga (for commutative differential graded algebra) free as a commutative graded algebra (∧V, d) and a morphism ϕ : (∧V, d) → ADR (M) inducing an isomorphism in cohomology. The first question is, can one build such a model for any manifold? The answer is yes for connected manifolds and in fact, there are many ways to do this. So, we have to define a standard way, which we call minimal. With this in mind, we again look to Sullivan’s introduction: One proceeds degree by degree to construct a smallest possible sub-differential algebra of forms with the same cohomology. Forms are chosen in each degree to add cohomology not already achieved or to create necessary relations among cohomology classes.

Once we have this minimal model, we may ask what geometrical invariants can be detected in it. In fact, there is a functor from algebra to geometry that, together with forms, creates a dictionary between the algebraic and the geometrical worlds. But for this we have to work over the rationals and not over the reals. As a consequence, we have to replace the de Rham algebra by other types of forms. At first glance, this seems to be a disadvantage because we are switching from a well-known object to an unfamiliar one. But this new construction is very similar to the de Rham algebra and will allow the extension of the usual theory from manifolds to topological spaces, which is a great advantage. Denote by APL (X) this analogue of the de Rham algebra for a topological space X. Since the minimal model construction also works perfectly well over Q, we have the notion of a minimal model MX → APL (X) of a path connected space X. Conversely, from a cdga (A, d) we have a topological realization (A, d) which is the return to Topology we hinted at above. If we apply this realization to a minimal model MX of a space X (which is nilpotent with finite Betti numbers), then we get a continuous map X → MX which induces an isomorphism in rational cohomology. The space MX is what, in homotopy theory, is called a rationalization of X. What must be emphasized in this process is the ability to create topological realizations of any algebraic constructions. Such a theory begs for applications and examples and we describe models for spheres, homogeneous spaces, biquotients, connected sums,

Preface

nilmanifolds, mapping spaces, configuration spaces and subspace arrangements. We give geometrical applications in several directions: to complex and symplectic manifolds, the closed geodesic problem, curvature problems, actions of tori, complements of submanifolds, symplectic blow-ups, and the Chas–Sullivan product, for instance. Roughly, this book is composed of three parts. The first part, consisting of Chapters 1–3, contains the classical theory and the main geometrical examples. These chapters are self-contained except for certain proofs for which we provide references. Chapters 4–8 are the second part. Each of them is devoted to a particular topic in differential topology or geometry and they are mostly independent. The third part is the florilège of Chapter 9 where we give short presentations of particular subjects, chosen to illustrate the evolution of applications of minimal models from the theory’s inception to the present day. Evidently we have been obliged to make choices in these applications and, therefore, many other interesting applications of algebraic models are not covered. The following brief description of the material in each of the chapters makes the outline above more precise. • Chapter 1. Throughout this book, Lie groups and homogeneous spaces

are used to give foundational examples and to show that some of the basic ideas of Sullivan’s rational homotopy theory were already present in this particular case years earlier. As well as describing certain basic structure results about Lie groups, this chapter gives a complete treatment of the computation of the cohomology algebra of a compact connected Lie group and recalls the basic facts about homogeneous spaces. We also look at the Cartan–Weil model and see it as the prototype for models of fibrations. • Chapter 2 is concerned with the basic definitions and properties of our algebraic tools: cdga’s, models, minimal models, homotopy between morphisms of cdga’s and the link between topological spaces and cdga’s. When we construct a minimal model for a cdga (A, d), it is possible that we do not have to consider the whole algebra of forms, but rather only the cohomology H(A, d). Although this is not true in general, it is true for spheres and Lie groups. This leads us to distinguish special types of spaces, called formal spaces, whose minimal models are determined by cohomology alone. This notion will be of great importance in applications and we delineate its properties. • Chapter 3. Since the main theme of this book is the geometrical aspect of algebraic models, a first question is, how special is a minimal model of a compact simply connected manifold? At the very least, its cohomology must satisfy Poincaré duality. In fact, it was proved recently that there is a

ix

x

Preface

model for X (and not just its cohomology) that satisfies Poincaré duality. What about the converse? How can we detect that the realization of a model contains a manifold in its rational homotopy type? Surprisingly, the conditions that are necessary for this also prove to be sufficient. Formality in the case of manifolds entails certain properties. We prove here the theorems of Miller and Stasheff giving particular instances when manifolds are formal. Notably, we show, as Stasheff did using another method, that a compact simply connected manifold M is formal if and only if M\{∗} is formal. We also extend the construction of models to the case of cdga’s equipped with the action of a finite group and apply it to the explicit construction of models of homogeneous spaces and biquotients. • In Chapter 4, we study the link between the Dolbeault and de Rham algebras of a complex manifold M as well as the relationship between the respective models. We carefully consider the topological consequences of the existence of a Kähler metric on M, and, in particular, we prove the formality of compact Kähler manifolds. We also consider the Dolbeault model of a complex manifold in detail and compute it in many particular cases, including the case of Calabi-Eckmann manifolds. For that, we use a perturbation theorem which allows the construction of a model of a filtered cdga starting from a model of any stage of the associated spectral sequence. Applications to the Frölicher spectral sequence of a complex manifold are given. In the last part of this chapter, we describe some of the implications of models for symplectic topology. For a compact symplectic manifold, we compare the hard Lefschetz property with other properties that appear in the complex situation. In particular, we recall results of Mathieu and Merkulov concerning the relation of the hard Lefschetz property to the existence of symplectically harmonic forms, as defined by Brylinski, in each cohomology class. • Chapter 5. For a smooth Riemannian manifold, the Riemannian structure of the manifold is reflected in its geodesics. The geodesics on a manifold may be viewed as the motion of a physical system, so in some sense, the study of geodesics exemplifies the paradigm expressing the relationship between mathematics and physics. Of course, the motions that are most important in physics are the periodic ones, so we begin by studying the geometric counterpart, closed geodesics. The main problem in this area is then: does every compact Riemannian manifold M of dimension at least two admit infinitely many geometrically distinct geodesics? The solution to this problem involves an essential analysis of the rational homotopy type of the free loop space of the manifold.

Preface

We give a minimal model of the free loop space and then prove the Vigué– Poirrier–Sullivan theorem, which solves most cases of the closed geodesic problem. We also present several other connections between the structure of models and properties of a manifold’s geodesics. Information about geodesics can often be codified by the dynamical system known as the geodesic flow and we shall see that the flow also holds rational homotopy information within it. • Chapter 6. In the last decade, algebraic models have proven to be useful tools in the study of various differential geometric questions involving curvature. A basic problem is whether curvature and diameter constraints limit to a finite number the possible rational homotopy types of manifolds satisfying those constraints. We describe the use of models in the construction of counterexamples to this question. We also show how models can be used to give a general analysis of the failure of the converse of the Soul theorem of Cheeger and Gromoll. • Chapter 7. The topological qualities of a space are often reflected in its intrinsic symmetries. These symmetries, in turn, may be formalized as the actions of groups on the space. Intuitively, most manifolds are asymmetric, so the existence of a nontrivial group action on a manifold implies that the manifold is special topologically. The properties of a manifold with group action may be gleaned from various topological constructions and their algebraic reflections. Indeed, this chapter focuses on what can be said about group actions from the viewpoint of algebraic models. For instance, there is a longstanding conjecture called the toral rank conjecture which is usually attributed to S. Halperin. The conjecture says, in particular, that if there is a free action of a torus T r on a space X, then the dimension of the rational cohomology of X must be at least as large as the dimension of the rational cohomology of the torus. We give proofs for homogeneous spaces and Kähler manifolds. We also discuss the Borel localization theorem and apply it to the study of the rational homotopy and the rational cohomology of fixed point sets. Finally, we discuss the notion of Hamiltonian action in symplectic geometry and use models to prove a special case of the Lalonde–McDuff question about Hamiltonian bundles. • Chapter 8. The process of taking a blow-up has proven to be extremely useful in complex and symplectic geometry. In order to consider various questions on the interface between geometry and algebraic topology, it is necessary to understand algebraic models of blow-ups. This entails a panoply of related questions which all serve as testaments to the efficacy of rational homotopy theory in geometry. In this chapter we consider two types of questions.

xi

xii

Preface

First of all, let f : N → M be a closed submanifold of a compact orientable manifold and denote by C its complement. The natural problem is to know if the rational homotopy type of C is completely determined by the rational homotopy type of the embedding, and in that case to describe a model for the injection C → M from a model of the initial embedding. We use it to describe nonformal simply connected symplectic manifolds that are blow-ups. Our second question concerns the geometric intersection theory of cycles in a compact manifold. M. Chas and D. Sullivan have extended the standard intersection theory to an intersection theory of cycles in the free loop space LN for any compact oriented manifold N. More precisely, they define a product on H∗ (LN) that combines the intersection product on the chains on N and the Pontryagin composition of loops in N. We present a more homological re-interpretation of the Chas–Sullivan product and, as a corollary, obtain the well-known theorem of Cohen and Jones. • In Chapter 9, we consider various types of geometric situations where algebraic models are useful. Models make their presence felt in the study of configuration spaces, arrangements, smooth algebraic varieties, mapping spaces, Gelfand–Fuchs cohomology, and iterated integrals. Of course, it is impossible to prove everything about such an array of topics, so this chapter is simply a survey of these applications of models. We endeavor to describe and explain the relevant models and then refer to the appropriate literature for details. • Finally, there are three appendices that recall basic facts about de Rham forms, spectral sequences and homotopy theory. It is our hope (and we believe) that this book will prove enlightening to both geometers and topologists. It should be useful to geometers because of concrete examples showing how algebraic techniques can be used to help solve geometric problems. For topologists, on the other hand, it is important to see what kind of concrete geometrical questions can be studied from a topological point of view. A project such as this requires a great deal of support and we would like to acknowledge this here. First, this book would never have seen the light of day without Research in Pairs grants from the Mathematisches Forschungsinstitut Oberwolfach in 2003 and 2006. These stays at the MFO were essential to our collaboration and it is a pleasure to acknowledge the generosity of this mathematical haven. Various portions of the book were read by Agusti Roig and he provided many insightful comments and suggestions. We also thank P. Lambrechts and G. Paternain for discussions

Preface

on several topics. Finally, the support of the University of Louvain-LaNeuve and of the CNRS for the Summer School on Algebraic Models, held at Louvain-La-Neuve in June 2007, was essential to the completion of this work. Let’s now begin.

xiii


Contents

vii

Preface

1

Lie groups and homogeneous spaces 1.1 1.2 1.3 1.4 1.5

1.6 1.7 1.8 1.9 1.10 1.11 1.12 1.13 2

Lie groups Lie algebras Lie groups and Lie algebras Abelian Lie groups Classical examples of Lie groups 1.5.1 Subgroups of the real linear group 1.5.2 Subgroups of the complex linear group 1.5.3 Subgroups of the quaternionic linear group Invariant forms Cohomology of Lie groups Simple and semisimple compact connected Lie groups Homogeneous spaces Principal bundles Classifying spaces of Lie groups Stiefel and Grassmann manifolds The Cartan–Weil model

1 2 3 5 8 8 9 10 10 11 16 21 26 32 38 42 47

Minimal models

56

2.1 2.2 2.3

57 61 64 64 66 67 67 69

2.4

Commutative differential graded algebras Homotopy between morphisms of cdga’s Models in algebra 2.3.1 Minimal models of cdga’s and morphisms 2.3.2 Relative minimal models Models of spaces 2.4.1 Real and rational minimal models 2.4.2 Construction of APL (X)

xvi

Contents

2.5

2.6

2.7

2.8 3

71 74 75 75 78 84 87 90 90 91 92 95 96 100

Manifolds

104

3.1

105 105 106 109 110 116 117 121 123 123 127 133 133 137 139

3.2

3.3

3.4

3.5 4

2.4.3 Examples of minimal models of spaces 2.4.4 Other models for spaces Minimal models and homotopy theory 2.5.1 Minimal models and homotopy groups 2.5.2 Relative minimal model of a fibration 2.5.3 The dichotomy theorem 2.5.4 Minimal models and some homotopy constructions Realizing minimal cdga’s as spaces 2.6.1 Topological realization of a minimal cdga 2.6.2 The cochains on a graded Lie algebra Formality 2.7.1 Bigraded model 2.7.2 Obstructions to formality Semifree models

Minimal models and manifolds 3.1.1 Sullivan–Barge classification 3.1.2 The rational homotopy groups of a manifold 3.1.3 Poincaré duality models 3.1.4 Formality of manifolds Nilmanifolds 3.2.1 Relations with Lie algebras 3.2.2 Relations with principal bundles Finite group actions 3.3.1 An equivariant model for -spaces 3.3.2 Weyl group and cohomology of BG Biquotients 3.4.1 Definitions and properties 3.4.2 Models of biquotients The canonical model of a Riemannian manifold

Complex and symplectic manifolds

145

4.1

148 148 150

Complex and almost complex manifolds 4.1.1 Complex manifolds 4.1.2 Almost complex manifolds 4.1.3 Differential forms on an almost complex manifold 4.1.4 Integrability of almost complex manifolds

152 154

Contents

4.2

4.3

4.4

4.5

4.6

4.7

5

Kähler manifolds 4.2.1 Definitions and properties 4.2.2 Examples: Calabi–Eckmann manifolds 4.2.3 Topology of compact Kähler manifolds The Dolbeault model of a complex manifold 4.3.1 Definition and existence 4.3.2 The Dolbeault model of a Kähler manifold 4.3.3 The Borel spectral sequence 4.3.4 The Dolbeault model of Calabi–Eckmann manifolds The Frölicher spectral sequence 4.4.1 Definition and properties 4.4.2 Pittie’s examples Symplectic manifolds 4.5.1 Definition of symplectic manifold 4.5.2 Examples of symplectic manifolds 4.5.3 Symplectic manifolds and the hard Lefschetz property 4.5.4 Symplectic and complex manifolds Cohomologically symplectic manifolds 4.6.1 C-symplectic manifolds 4.6.2 Symplectic homogeneous spaces and biquotients 4.6.3 Symplectic fibrations 4.6.4 Symplectic nilmanifolds 4.6.5 Homotopy of nilpotent symplectic manifolds Appendix: Complex and symplectic linear algebra 4.7.1 Complex structure on a real vector space 4.7.2 Complexification of a complex structure 4.7.3 Hermitian products 4.7.4 Symplectic linear algebra 4.7.5 Symplectic and complex linear algebra 4.7.6 Generalized complex structure

156 156 159 162 168 169 172 173 175 178 178 179 182 182 183 184 187 187 187 188 189 191 194 196 196 197 198 200 201 202

Geodesics

205

5.1 5.2 5.3 5.4 5.5

207 210 213 215 222

The closed geodesic problem A model for the free loop space A solution to the closed geodesic problem A-invariant closed geodesics Existence of infinitely many A-invariant geodesics

xvii

xviii

Contents

5.6 5.7 5.8 5.9 6

223 227 232 234

Curvature

239

6.1 6.2

239 243 243 249 252 253 259 265 266

6.3

6.4 6.5 7

Gromov’s estimate and the growth of closed geodesics The topological entropy Manifolds whose geodesics are closed Bar construction, Hochschild homology and cohomology

Introduction: Recollections on curvature Grove’s question 6.2.1 The Fang–Rong approach 6.2.2 Totaro’s approach Vampiric vector bundles 6.3.1 The examples of Özaydin and Walschap 6.3.2 The method of Belegradek and Kapovitch Final thoughts Appendix

G-spaces

271

7.1 7.2 7.3

273 275 276 278 280 283 287 289 291 292 295 298 298 303 305 306 306 308 312 312 314

7.4

7.5

7.6

Basic definitions and results The Borel fibration The toral rank 7.3.1 Toral rank for rationally elliptic spaces 7.3.2 Computation of rk0 (M) with minimal models 7.3.3 The toral rank conjecture 7.3.4 Toral rank and center of π∗ (M) ⊗ Q 7.3.5 The TRC for Lie algebras The localization theorem 7.4.1 Relations between G-manifold and fixed set 7.4.2 Some examples The rational homotopy of a fixed point set component 7.5.1 The rational homotopy groups of a component 7.5.2 Presentation of the Lie algebra LF = π∗ (F) ⊗ Q 7.5.3 Z/2Z-Sullivan models Hamiltonian actions and bundles 7.6.1 Basic definitions and properties 7.6.2 Hamiltonian and cohomologically free actions 7.6.3 The symplectic toral rank theorem 7.6.4 Some properties of Hamiltonian actions 7.6.5 Hamiltonian bundles

Contents

8

Blow-ups and Intersection Products

317

8.1

318 319 321 324 328

8.2

8.3

8.4

9

The model of the complement of a submanifold 8.1.1 Shriek maps 8.1.2 Algebraic mapping cones 8.1.3 The model for the complement C 8.1.4 Properties of Poincaré duality models 8.1.5 The configuration space of two points in a manifold Symplectic blow-ups 8.2.1 Complex blow-ups 8.2.2 Blowing up along a submanifold A model for a symplectic blow-up 8.3.1 The basic pullback diagram of PL-forms 8.3.2 An illustrative example 8.3.3 The model for the blow-up 8.3.4 McDuff’s example 8.3.5 Effect of the symplectic form on the blow-up 8.3.6 Vanishing of Chern classes for KT The Chas-Sullivan loop product on loop space homology 8.4.1 The classical intersection product 8.4.2 The Chas–Sullivan loop product 8.4.3 A rational model for the loop product 8.4.4 Hochschild cohomology and Cohen–Jones theorem 8.4.5 The Chas-Sullivan loop product and closed geodesics

329 330 331 332 334 334 334 335 337 339 339 341 341 342 344 346 348

A Florilège of geometric applications

350

9.1

351 352 353 354 355 358

9.2

Configuration spaces 9.1.1 The Fadell–Neuwirth fibrations 9.1.2 The rational homotopy of configuration spaces 9.1.3 The configuration spaces F(Rn , k) 9.1.4 The configuration spaces of a projective manifold Arrangements 9.2.1 Formality of the complement of a geometric lattice 9.2.2 Rational hyperbolicity of the space M(A)

361 362

xix

xx

Contents

9.3 9.4 9.5

9.6

9.7

A

B

Toric topology Complex smooth algebraic varieties Spaces of sections and Gelfand–Fuchs cohomology 9.5.1 The Haefliger model for spaces of sections 9.5.2 The Bousfield–Peterson–Smith model 9.5.3 Configuration spaces and spaces of sections 9.5.4 Gelfand–Fuchs cohomology Iterated integrals 9.6.1 Definition of iterated integrals 9.6.2 The cdga of iterated integrals 9.6.3 Iterated integrals and the double bar construction 9.6.4 Iterated integrals, the Hochschild complex and the free loop space 9.6.5 Formal homology connection and holonomy 9.6.6 A topological application Cohomological conjectures 9.7.1 The toral rank conjecture 9.7.2 The Halperin conjecture 9.7.3 The Bott conjecture 9.7.4 The Gromov conjecture on LM 9.7.5 The Lalonde–McDuff question

363 364 367 367 371 373 375 376 376 379 381 384 385 387 388 388 388 389 390 390

De Rham forms

392

A.1 A.2 A.3 A.4

392 398 402 404

Differential forms Operators on forms The de Rham theorem The Hodge decomposition

Spectral sequences

409

B.1 B.2 B.3 B.4 B.5 B.6

409 411 412 413 416

B.7

What is a spectral sequence? Spectral sequences in cohomology Spectral sequences and filtrations Serre spectral sequence Zeeman–Moore theorem An algebraic example: The odd spectral sequence A particular case: A double complex

419 420

Contents

C

Basic homotopy recollections C.1 C.2 C.3

n-equivalences and homotopy sets Homotopy pushouts and pullbacks Cofibrations and fibrations

423 423 424 428

References

433

Index

451

xxi


1

Lie groups and homogeneous spaces

Lie groups and homogeneous spaces form an important family of examples of manifolds. We will use them systematically in many different parts of this book, to give an illustration for a specific method or, as well, to show that new viewpoints can be obtained by using algebraic models. Therefore, it is important for us to understand basic ideas and results about Lie groups in order to appreciate the development and application of algebraic models to geometry in general. What we also want to articulate in the next several chapters is that some of the basic ideas of Sullivan’s rational homotopy theory were already present in this particular case. Of course, the (rational or real) cohomology algebra is a fundamental example of an algebraic model, but it was understood from the start that only certain information was reflected in it. Nevertheless, computing cohomology for such geometric building blocks as Lie groups became an important goal in the early years of algebraic topology. An algebraic model that proved to be effective in reaching this goal was the Cartan–Weil model. Nowadays, we look at this model and see it as the direct ancestor of Sullivan’s minimal models. Indeed, the Cartan–Weil model is the prototype for models of fibrations (see Chapter 2) and, thus holds within it homotopical, rather than just cohomological, information. A complete treatment of Lie group theory covers several books, so we need to make precise the philosophy of this chapter. Basic properties and definitions on Lie groups and Lie algebras are recalled without proofs in Sections 1.1–1.4. The classical examples of groups of matrices are described in Section 1.5. From that point on, we wish to be complete, with proofs, when we discuss the things that form the main focus of the book: here, that means in particular, the de Rham cohomology of Lie groups and homogeneous spaces. Moreover, and we will discover the importance of this in Chapter 2, we wish to have a “computable” cochain complex linked to the de Rham complex by an algebraic map, → ADR , which induces an isomorphism in cohomology. For instance, in the case of Lie groups, we prove in Section 1.6 that the inclusion I (G) → ADR (G) of the bi-invariant

2

1 : Lie groups and homogeneous spaces

forms induces an isomorphism in cohomology. We continue in Section 1.7 by providing two structure theorems for the cohomology of Lie groups, including the Hopf theorem which expresses the cohomology algebra as an exterior algebra on generators of odd degrees. We will use this result in Section 1.8 for the computation of the second and third Betti numbers for simple Lie groups. Since fibrations are the basic ingredients in the theory of minimal models, we develop the notions of principal bundles and classifying spaces. Sections 1.9–1.10 contain definitions of bundles, principal bundles and homogeneous spaces. In Section 1.11, we define and characterize the notion of classifying space for Lie groups. The classical Stiefel and Grassmann manifolds provide us with classifying spaces for the various orthogonal groups and they are studied in Section 1.12. Finally, in Section 1.13, we give the prototype for the theory of minimal models (see Chapter 2), the Cartan–Weil model for equivariant cohomology. We assume that the reader possesses a good knowledge of the basic concepts about manifolds, as in [28], [104], [226], or (for surface theory) [215] for instance. We also assume the reader has had a classical homology course. Appendices A, B and C supply the basic recollections on some crucial parts of these subjects. In this chapter, manifold means a Hausdorff space with a countable basis of open sets (i.e. a separable space), endowed with a differentiable structure of class C ∞ over the reals.

1.1

Lie groups

In this section, we give the definitions and basic properties of Lie groups. General references are [2], [136] and [199] for instance. Definition 1.1 A Lie group is a set G which is both a manifold and a group and for which the multiplication, (g, g ) → gg , and the inverse map, g → g −1 , are smooth. The dimension of a Lie group is its dimension as a manifold. A homomorphism of Lie groups is a homomorphism of groups which is also a smooth map. An isomorphism of Lie groups is a homomorphism f which admits an inverse f −1 as maps and such that f −1 is also a homomorphism of Lie groups. For instance, R and S1 are Lie groups for the usual structures of manifolds and groups. One can observe directly from the definition that a product of two Lie groups is a Lie group for the two canonical structures, product of groups and product of manifolds. So Rn and (S1 )n are Lie groups. The tori

1.2

Lie algebras

T n = (S1 )n are studied in Section 1.4, and Section 1.5 is devoted to classical examples of Lie groups, such as the different groups of matrices. Observe now that, if V is a real vector space, the set Gl(V) of linear isomorphisms is a Lie group. Remark 1.2 Some properties required in Definition 1.1 are in fact automatic: • if the multiplication is a smooth map, then the inverse is also a smooth

map (use the implicit function theorem); • a bijective homomorphism of Lie groups is an isomorphism of Lie groups

(see [214, page 18]); • a map f : G → H between Lie groups which is a homomorphism of

groups and a continuous map is a homomorphism of Lie groups (see [199, page 44]). A Lie subgroup of a Lie group G is a subgroup which is a submanifold of G. Lie subgroups can be determined easily by the following theorem of Elie Cartan. Theorem 1.3 ([2, page 17], [214, page 47]) A subgroup H of a Lie group G is a Lie subgroup of G if and only if H is a closed subgroup of G. In this book, we will be especially interested in the homotopy types of Lie groups and, for that, we can reduce the study to compact connected Lie groups as shown by the following result called the polar decomposition or Iwasawa decomposition. Theorem 1.4 ([147]) Any connected Lie group G admits a maximal compact subgroup H (unique up to conjugacy) such that G is isomorphic to the product H × Rm . In particular G and H have the same homotopy type. Compact connected Lie groups will be classified in Theorem 1.52.

1.2

Lie algebras

We introduce here the notion of Lie algebras and the example of main interest for us, the tangent space Te (G) of a Lie group G at the identity. Definition 1.5 A Lie algebra over R is a vector space l together with a bilinear homomorphism, called the bracket, [−, −] : l × l → l

3

4


such that, for any l1 ∈ l, l2 ∈ l, l3 ∈ l, one has: • [l1 , l2 ] = −[l2 , l1 ] (skew symmetry), • [l1 , [l2 , l3 ]] + [l2 , [l3 , l1 ]] + [l3 , [l1 , l2 ]] = 0 (Jacobi identity).

A homomorphism of Lie algebras is a linear map, ϕ : l → l , preserving the bracket. This means that ϕ[l1 , l2 ] = [ϕ(l1 ), ϕ(l2 )] for any (l1 , l2 ) ∈ l × l. A Lie subalgebra of a Lie algebra l is a sub-vector space n such that [n, n] ⊆ n. An ideal of l is a Lie subalgebra n such that [n, l] ⊆ n. Any structure of associative algebra on a vector space A gives a canonical structure of Lie algebra lA on the same vector space by [a1 , a2 ] = a1 a2 −a2 a1 . Our most interesting example comes from the structure of a Lie group G. Observe that, because of the group structure, any phenomenon at a particular point of G can be translated everywhere in G by composing with the elements of the group. For instance, a connected Lie group is generated, as a topological space, by any neighborhood of the identity e. We formalize this remark by introducing the notion of left and right translations. Definition 1.6 A fixed element g ∈ G gives the left translation Lg : G → G with Lg (h) = g · h for all h ∈ G. Similarly, we define right translations Rg by Rg (h) = h · g. Recall first that, if p : T(G) → G is the tangent bundle of the manifold G, a vector field X on G is a smooth section of p (see Appendix A). Definition 1.7 Denote by DLg : T(G) → T(G) the map induced by the left translation Lg . A vector field X on G is called left invariant if DLg (X) = X, for any g ∈ G. Remark 1.8 Left invariance of objects other than vector fields also turns out to be very important in understanding Lie groups. In particular, if we define a function f on a Lie group G to be left invariant when f (gh) = f (h) for all g, h ∈ G, then clearly we see that f is a constant function. Later (see Definition 1.25), we will phrase this by saying that any left invariant 0-form is a constant. If G is a Lie group, we denote by g the vector space of left invariant vector fields on G. If X and Y are vector fields, then their bracket is defined to be the vector field [X, Y]f = X(Yf ) − Y(Xf ) for all functions f . The bracket is anti-commutative and satisfies a Jacobi identity (see Section A.2). If X and Y are left invariant vector fields, their bracket [X, Y] is also left invariant. Therefore, the vector space g has the structure of a Lie algebra, called the Lie algebra associated to the Lie group G.

1.3

Lie groups and Lie algebras

If X is a vector field on a Lie group G, we see directly from the definition that X is left invariant if and only if Xg = (DLg )(Xe ). Therefore the vector space g is isomorphic to the tangent space, Te (G), at the identity e of G. We will not make any distinction between these two characterizations of the Lie algebra g. From this observation, we deduce that the left invariant vector fields provide n linearly independent sections of the tangent bundle TG for an n-dimensional Lie group G. Therefore, any Lie group G is parallelizable; that is, TG ∼ = G × Te (G) = G × g. As a consequence, the real Lie algebra of vector fields on G (see Section A.2) is the tensor product C∞ (G) ⊗ g where C∞ (G) is the algebra of smooth real valued functions on G. Proposition 1.9 There is a morphism of Lie groups Ad : G → Gl(g) given by Ad(g)(X) = ((DRg )−1 ◦ (DLg ))(X), where Gl(g) is the group of linear isomorphisms of the Lie algebra g. Definition 1.10 The map Ad : G → Gl(g) is called the adjoint representation of the Lie group G. Proposition 1.11 Denote by gl(g) the Lie algebra of the Lie group Gl(g). Then, the derivative of Ad : G → Gl(g) is the morphism of Lie algebras, ad : g → gl(g), defined by ad(X)(Y) = [X, Y], where [−, −] is the bracket of g.

1.3


It is now time to make various relations between Lie groups and Lie algebras precise. A basic reference is [214, Section I-&2]. First, by definition of the Lie algebra associated to a Lie group, if f : G → H is a homomorphism of Lie groups, its differential Dfe : g → h is a homomorphism of Lie algebras. We would like to know if Dfe gives us information about the homomorphism f . For that, we need a better correspondence between Lie algebras and Lie groups which comes from the notion of one-parameter subgroup. Definition 1.12 For any Lie group G, a homomorphism of Lie groups, θ : R → G, is called a one-parameter subgroup of G. Observe that, by definition, such a θ satisfies θ (s + t) = θ (s) · θ (t), for any s and t in R. It can be shown that any one-parameter subgroup is the integral

5

6


curve of a left invariant vector field and, reciprocally, that any left invariant vector field admits a one-parameter subgroup as a maximal integral curve. So, we get a new characterization of the Lie algebra associated to a Lie group. Theorem 1.13 There is an isomorphism between the Lie algebra g associated to a Lie group G and the set of one-parameter subgroups of G. With this isomorphism, one can construct a map from g to G. Definition 1.14 If X ∈ g, we denote by θX the one-parameter subgroup associated to X as in Theorem 1.13. The exponential from g = Te (G) to G is defined by exp(tX) = θX (t). Observe that, by uniqueness of the integral curve, one has θλX (t) = θX (λt) and the exponential is well defined. In fact, the exponential is a smooth map which induces the identity on the tangent space at e ∈ G; that is, Dexp = id : g → g. It can also be shown that the exponential is an epimorphism if the group G is compact and connected. Moreover, if f : G → G is a homomorphism of Lie groups, then one has f ◦ expG = expG ◦ Df . This process of integration is the key for the two next results. The first one concerns the link between a homomorphism of Lie groups and its induced differential. Theorem 1.15 Let G and H be two Lie groups with G connected. Then a homomorphism from G to H is uniquely determined by its differential Dfe : g → h. The second result concerns the realization of morphisms between Lie algebras. Theorem 1.16 Let G and H be two Lie groups with G simply connected. Then, for every homomorphism of Lie algebras ψ : g → h, there exists a homomorphism of Lie groups f : G → H such that Dfe = ψ. The theory is very powerful. For instance, the third Lie theorem gives a converse to the construction of the Lie algebra of a Lie group (see [229, Leçon 6]). Theorem 1.17 Every finite dimensional Lie algebra is the tangent space algebra of some Lie group.

1.3


This association can be made more precise: • every finite dimensional Lie algebra is the tangent space algebra of a

unique simply connected Lie group;

• if a homomorphism of Lie groups, f : G → G , with G simply connected,

induces an isomorphism between the associated Lie algebras, then f is a universal cover. These results imply a correspondence between sub-Lie groups and subLie algebras which can be made explicit for some objects of interest. For the rest of this section, let G be a connected Lie group with associated Lie algebra g and let H be a Lie subgroup of G. Then the Lie algebra h associated to H is a Lie subalgebra of g and the subgroup H is normal if and only if the subalgebra h is an ideal. Now let’s recall some classical definitions (see [113, page 69]) which will be useful in the rest of this section. Definition 1.18 The centralizer of a subset A of G is the subgroup Z(A) = x ∈ G | xa = ax for any a ∈ A . The centralizer of G is called the center of G. The centralizer of a subset m in g is the Lie subalgebra Z (m) = l ∈ g | [l, m] = 0 for any m ∈ m . The centralizer of g is called the center of g. It can be shown that the centralizer of H in G is a Lie group with associated Lie algebra the centralizer of h in g. Definition 1.19 The normalizer of a subset A of G is the subgroup of G given by N(A) = {x ∈ G | xA = Ax} . The normalizer of a subset m of g is the Lie subalgebra n(m) = x ∈ g | [x, y] ∈ m for any y ∈ m . It can be shown that the normalizer of H in G is a Lie group with associated Lie algebra the normalizer of h in g. Observe that, for any x ∈ N(H) and any h ∈ H, we have xhx−1 ∈ H. We deduce that there is an action (see Definition 1.23) of the Lie group N(H) on the manifold H, called the conjugation action.

7

8


1.4

Abelian Lie groups

It turns out that it is essential to first study the abelian Lie groups and the abelian Lie subgroups of a compact connected Lie group. A reference for this section is [113, Chapters 1 and 2]. A basic example of an abelian compact Lie group is the circle S1 endowed with the commutative multiplication of complex numbers. More generally, we have the Definition 1.20 An abelian Lie group is a Lie group G satisfying gg = g g for any (g, g ) ∈ G × G. An abelian Lie algebra is a Lie algebra l such that [l, l ] = 0 for any (l, l ) ∈ l × l. One can prove that a Lie group G is abelian if and only if its Lie algebra g is abelian. A product of n circles is an abelian Lie group, called an n-torus (or simply a torus) and denoted by T n . Tori are the prototypes of abelian Lie groups. Theorem 1.21 Any connected abelian Lie group G is isomorphic to the direct product of Lie groups T p × Rq . As a consequence, any connected abelian Lie subgroup of a compact connected Lie group G is a torus T. Call a subtorus T ⊂ G a maximal torus in G if it is not properly contained in another torus. One can then prove the following. Theorem 1.22 Every element of a compact connected Lie group is contained in a maximal torus. Two maximal tori are conjugate. The dimension of a maximal torus is called the rank of the Lie group. The normalizer of a maximal torus T of G is a compact Lie group denoted N(T). The quotient W(G) = N(T)/T is called the Weyl group of G. Up to isomorphism, this group does not depend on the choice of a maximal torus in G. In fact, W(G) is a finite group. Observe that, since T is abelian, the restriction of the conjugation action of N(T) to T is trivial on T, so it gives an action of W(G) on T. We will come back to the study of maximal tori in Subsection 3.3.2.

1.5

Classical examples of Lie groups

We now describe the classical examples of Lie groups which come from groups of matrices. A reference for this section is [199, Chapter I].

1.5

1.5.1

Classical examples of Lie groups

Subgroups of the real linear group

Start with the set of real numbers R and denote by Gl(n, R) the group of invertible n × n-matrices with entries in R. Endowed with the canonical 2 structure of a manifold (as an open subset of Rn ), the group Gl(n, R) is a Lie group, called the real linear Lie group. The associated Lie algebra, gl n (R) = M(n, R), is the vector space of all n × n matrices with the bracket being the commutator of matrices. The dimension of Gl(n, R) is n2 . Any closed subgroup of Gl(n, R) is a Lie group. In particular, we have the orthogonal group O(n) consisting of the orthogonal linear transformations u of the euclidian space Rn . Recall that, in the canonical basis, this is equivalent to the fact that the matrix A of u satisfies t AA = In . The associated Lie algebra o(n) of O(n) is the vector space of alternating (or skew-symmetric) n(n − 1) n t matrices, A = −A. The dimension of O(n) is = and O(n) 2 2 is a maximal compact subgroup of Gl(n, R). Since the continuous map det : O(n) → {−1, +1} is surjective, one sees that the space O(n) is not connected. We denote by SO(n) the subgroup of O(n), consisting of linear transformations of determinant 1 and call it the special orthogonal group. Since it is a connected component of O(n), the group SO(n) has the same tangent space at the neutral element e, therefore the same Lie algebra, by definition. As we will see, the group SO(n) is not simply connected if n ≥ 2. The universal cover of SO(n) is called the nth-spinor group and denoted by Spin(n). The orthogonal group O(n) is the prototype of Lie groups. Indeed, it can be proved that any compact Lie group G is isomorphic to a closed subgroup of O(n) (see [199, Theorem 2.14, Chapter V]). If θ is a real number, we denote by cos θ − sin θ R(θ) = sin θ cos θ the rotation matrix corresponding to the rotation in R2 by angle θ . Let (θ1 , . . . , θr ) be r real numbers. Denote by (R(θ1 ), . . . , R(θr ), 1) the matrix having the R(θi ) and 1 along the diagonal and 0 entries otherwise. The group of matrices of the form (R(θ1 ), . . . , R(θr ), 1) is the maximal torus of SO(2r + 1) and the rank of SO(2r + 1) is r. The Weyl group of SO(2r + 1) has 2r r! elements and acts on the maximal torus by a permutation of the coordinates composed with the substitutions (θ1 , . . . , θr ) → (±θ1 , . . . , ±θr ). As for the Lie group SO(2r), its maximal torus consists of matrices (R(θ1 ), . . . , R(θr )) and the rank of SO(2r) is r. Its Weyl group has 2r−1 r! elements acting on the maximal torus by a permutation of the coordinates

9

10


composed with the substitutions (θ1 , . . . , θr ) → (ε1 θ1 , . . . , εr θr ), with εi = ±1 and ε1 · · · εr = 1.

1.5.2

Subgroups of the complex linear group

Denote by Gl(n, C) the group of invertible n × n-matrices with entries in the complex numbers C. Endowed with the canonical structure of a manifold 2 (as an open subset of R2n ), the group Gl(n, C) is a Lie group, called the complex linear Lie group. The associated Lie algebra, gl n (C) = M(n, C), is the vector space of all n×n-matrices with the bracket being the commutator of matrices. The (real) dimension of Gl(n, C) is 2n2 . We now introduce the analogue of the orthogonal group. Recall that, if we write a complex number as z = x + iy, with x ∈ R and y ∈ R, the conjugate √ of z is the complex number z = x − iy. This induces a norm with z = zz. The unitary group U(n) consists of the linear transformations u of R2n that respect this norm; that is, u(z) = z. In the canonical basis, this is equivalent to the fact that the matrix A of u satisfies t AA = In . The associated Lie algebra u(n) of U(n) is the vector space of alternating (or skew) hermitian matrices, t A = −A. The dimension of U(n) is n2 . It can be proved that U(n) is a maximal compact subgroup of Gl(n, C) and that U(n) = SO(2n) ∩ Gl(n, C). The subgroup of U(n) consisting of linear transformations of determinant 1 is called the special unitary group and denoted by SU(n). The associated Lie algebra, su(n), consists of matrices of trace 0 such that t A = −A. The dimension of SU(n) is n2 − 1. The group SU(n) is simply connected. The group U(n) is not, but its universal cover does not constitute something new because, as a space, U(n) is diffeomorphic to the product S1 × SU(n). The maximal torus of U(n) consists of the set of diagonal matrices having (eiλ1 , . . . , eiλn ) on the diagonal. The Lie group U(n) has rank n. Its Weyl group is the symmetric group n acting on the maximal torus by a permutation of the coordinates. The maximal torus of SU(n) consists of the set of diagonal matrices having (eiλ1 , . . . , eiλn ) on the diagonal such that ni=1 λi = 0. The Lie group SU(n) has rank n − 1. The Weyl group and its action are the same as for U(n).

1.5.3

Subgroups of the quaternionic linear group

Now consider the field of quaternions H and denote by Gl(n, H) the group of invertible n × n-matrices with entries in H. Endowed with the canonical 2 structure of a manifold (as an open subset of R4n ), the group Gl(n, H) is a Lie group, called the quaternionic linear Lie group. The associated Lie algebra, gl n (H) = M(n, H), is the vector space of all n × n matrices with the

1.6

Invariant forms

bracket being the commutator of matrices. The (real) dimension of Gl(n, H) is 4n2 . For the associated orthogonal group, we have to define a quaternionic conjugation. Let z = t + ix + jy + kz be a quaternion, with x ∈ R, y ∈ R, z ∈ R, t ∈ R and i, j, k obeying the usual relations: i2 = j2 = k2 = −1, ij = k, ji = −k, jk = i, kj = −i, ki = j and ik = −j. The conjugate of z√is the quaternion z = t − ix − jy − kz. This induces a norm with z = zz. The symplectic group Sp(n) consists of the linear transformations u of R4n that respect this norm, u(z) = z. In the canonical basis, this is equivalent to the fact that the matrix A of u satisfies t AA = In . The associated Lie algebra sp(n) of Sp(n) is the vector space of alternating (or skew) quaternionic matrices, t A = −A. The dimension of Sp(n) is n(2n+1). One can show that Sp(n) is a maximal compact subgroup of Gl(n, H) and that Sp(n) = SO(4n) ∩ Gl(n, H). Viewed as a subgroup of U(2n) (see Exercise 1.3), the Lie group Sp(n) has for a maximal torus the diagonal matrices (eiλ1 , . . . , eiλ2n ) such that λi = λi+n for any 1 ≤ i ≤ n. The Lie group Sp(n) has rank n. Its Weyl group has 2n n! elements acting on the maximal torus as in SO(2n + 1). In any of these groups of matrices, the exponential map, exp : g → G, is the traditional exponential of a matrix: exp(A) = 1 + A + · · · +

1.6

An + ··· . n!

Invariant forms

In this section, we define the complex of invariant forms on a left Gmanifold M, and prove that the cohomology of this complex is isomorphic to the cohomology of M if the manifold M is compact and the Lie group G compact and connected. As we will see in several places, Lie groups are designed as groups of symmetries of manifolds. With this in mind, we define invariant forms in the general setting of G-manifolds. Definition 1.23 A Lie group G acts on a manifold M, on the left, if there is a smooth map G × M → M, (g, x) → gx, such that (g · g )x = g(g x) and ex = x for any x ∈ M, g ∈ G, g ∈ G. Such data endows M with the appellation of a left G-manifold. A left action is called • effective if gx = x for all x ∈ M implies g = e; • free if gx = x for any x ∈ M implies g = e.

For right actions and right G-manifolds, we ask for a smooth map M × G → M, (x, g) → xg, such that x(g ·g ) = (xg)g and xe = x for any x ∈ M, g ∈ G, g ∈ G.

11

12


Example 1.24 Let G be a Lie group. The Lie multiplication gives to G the structure of a • left G-manifold, with L : G × G → G, L(g, g ) = Lg (g ) = g · g ; • right G-manifold, with R : G × G → G, R(g , g) = Rg (g ) = g · g.

Let G be a Lie group. If M is a left G-manifold, we denote by g ∗ : ADR (M) → ADR (M) the “pullback” map induced on differential forms by the action of g ∈ G. More specifically, for vector fields X1 , . . . , Xk and a k-form ω, we define at m ∈ M, g ∗ ω(X1 , . . . , Xk )(m) = ωg·m (Dgm X1 (m), . . . , Dgm Xk (m)). We sometimes write ωx (X1 , . . . , Xk ) = ω(X1 , . . . , Xk )(x), L∗g ω = g ∗ ω and Dg = DLg . Definition 1.25 An invariant form on a left G-manifold M is a differential form ω ∈ ADR (M) such that g ∗ ω = ω for any g ∈ G. We denote the set of invariant forms by L (M). In the case of a Lie group G, we note that the left invariant forms (right invariant forms) correspond to the left (right) translation action. We denote these sets by L (G) and R (G) respectively. A form on G that is left and right invariant is called bi-invariant (or invariant if there is no confusion). The corresponding set is denoted by I (G). The aim of this section is to prove that these different sets of invariant forms allow the determination of the cohomology of G-manifolds and Lie groups. First, using the operators i(X) and L(X) on forms discussed in Appendix A (more specifically in Section A.2), we observe the following. Proposition 1.26 Let G be a Lie group and M be a left (or a right) G-manifold. Then the set of invariant forms of M is stable under d. Moreover, the sets of left invariant forms and of right invariant forms on G are invariant under i(X) and L(X), for X a left invariant vector field. Proof Suppose ω is a left invariant form on G and X is a left invariant vector field on G. We have, using the left invariance of X and ω, Lg ∗ i(X)ω(Y1 , . . . , Yk )(x) = Lg ∗ ω(X, Y1 , . . . , Yk )(x) = ωgx (DLg (X)x , DLg (Y1 )x , . . . , DLg (Yk )x ) = i(DLg X)ω(DLg (Y1 ), . . . , DLg (Yk ))(gx) = i(X)ω(Y1 , . . . , Yk )(x).

1.6

Invariant forms

Hence, i(X)ω is left invariant. The verification of the other statements is similar. The previous result justifies the following definition. Definition 1.27 Let G be a Lie group and M be a left G-manifold. The invariant cohomology of M is the homology of the cochain complex (L (M), d). We denote it by HL∗ (M). The main result is the following theorem. Theorem 1.28 Let G be a compact connected Lie group and M be a compact left G-manifold. Then HL∗ (M) ∼ = H ∗ (M; R). We will prove that the injection map L (M) → ADR (M) induces an isomorphism in cohomology. For that, we need some results concerning integration on a compact connected Lie group. Proposition 1.29 On a compact connected Lie group, there exists a biinvariant volume form. Proof Recall from Section 1.2 that the tangent bundle of G trivializes as T(G) ∼ = G × g. If g∗ is the dual vector space of g, we therefore have a trivialization of the cotangent bundle T ∗ (G) ∼ = G × g∗ and of the differential forms bundle. Exactly as for vector fields, we observe that left (right) invariant forms are totally determined by their value at the unit e and that we have isomorphisms L (G) ∼ = R (G) ∼ = ∧g∗ , where ∧g∗ is the exterior algebra on the vector space g∗ . To make this space precise, recall that the elements of g∗ are left invariant 1-forms dual to left invariant vector fields. If we choose a basis {ω1 , . . . , ωn } dual to a basis of left invariant vector fields, an element of ∧g∗ may be written α= ai1 ···ip ωi1 · · · ωip where the ai1 ···ip ’s are constant. Choose such an α of degree n equal to the dimension of G. We associate to α a unique left invariant form αL such that (αL )e = α and a unique right invariant form αR such that (αR )e = α. More precisely, we set: (αL )g (X1 , . . . , Xn ) = α((DLg )−1 X1 , . . . , (DLg )−1 Xn ), (αR )g (X1 , . . . , Xn ) = α((DRg )−1 X1 , . . . , (DRg )−1 Xn ).

13

14


Recall, from Definition 1.10, the homomorphism of Lie groups Ad : G → Gl(g). As direct consequences of the definitions, we have (L∗g αR )h (X1 , . . . , Xn ) = (αR )gh (DLg (X1 ), . . . , DLg (Xn )) = α((DRgh )−1 ◦ (DLg )(X1 ), . . .) = α((DRh )−1 ◦ (DRg )−1 ◦ DLg (X1 ), . . .) = (αR )h ((DRg )−1 ◦ DLg (X1 ), . . .) = (det(Ad(g))(αR )h (X1 , . . . , Xn ). The composition det ◦Ad : G → R has for image a compact subgroup of R; that is, {1} or {−1, 1}. Since the group G is connected, we get det(Ad(g)) = 1, for any g ∈ G, and αR is a bi-invariant volume form. The previous result can be obtained in a more general context. As the proof shows, it is sufficient to have (det ◦Ad)(g) = 1 for any g ∈ G. This is the definition of a unimodular group. Proof of Theorem 1.28 Denote by ι : L (M) → ADR (M) the canonical injection of the set of left invariant forms. We choose thebi-invariant volume form on G such that the total volume of G is 1, G dg = 1. This volume form allows the definition of G f dg ∈ Rk for any smooth function f : G → Rk . Let ω ∈ AkDR (M) and x ∈ M be fixed. As a function f , we take G → ∧Tx (M)∗ , g → g ∗ ω(x). We get a differential form ρ(ω) on M defined by: ρ(ω)(X1 , . . . , Xk )(x) = g ∗ ω(X1 , . . . , Xk )(x) dg G

=

G

(Lg )∗ ω(X1 , . . . , Xk )(x) dg.

We have thus built a map ρ : ADR (M) → ADR (M) and we now analyze its properties. Fact 1: ρ(ω) ∈ L (M).

Let g ∈ G be fixed. The map DLg : Tx (M) → Tg x (M) induces a map

∗ ∧ DLg : ∧ Tg x (M)∗ → ∧Tx (M)∗ . Therefore, one has (in convenient shorthand):

∗

∗ (Lg )∗ ω(x) dg DLg ρ(ω)(x) = ∧ DLg G

∗ = Lg ·g ω(x) dg G

1.6

=

G

Invariant forms

∗ Lg ω(x) dg

= ρ(ω)(x). Fact 2: If ω ∈ L (M) then ρ(ω) = ω. ∗ If Lg ω(x) = ω(x), then ρ(ω)(x) = G (Lg )∗ ω(x) dg = ω(x) G dg = ω(x). Fact 3: ρ ◦ d = d ◦ ρ. This is an easy verification from the definitions of d and ρ. From Facts 1–3, we deduce that H(ρ) ◦ H(ι) = id and H(ι) is injective. Fact 4: The integration can be reduced to a neighborhood of e. Let U be a neighborhood of e. We choose a smooth function ϕ : G → R, with compact support included in U, such that G ϕ dg = 1. Now we denote the bi-invariant volume form dg by ωvol . By classical differential calculus on manifolds, the replacement of ωvol by ϕ ωvol leaves the integral unchanged. The fact that ϕ ωvol has its support in U allows the reduction of the domain of integration to U. Our construction process can now be seen in the following light. ∗ (ϕω ) the Let L : G × M → M be the action of G on M. Denote by πG vol pullback of ϕωvol to ADR (U × M) by the projection πG : G × M → G and by L∗ : ADR (M) → ADR (U × M) the map induced by L. If α is a form ∗ (ϕω ) over the on U × M, we denote by I(α) the integration of α ∧ πG vol U-variables, considering the variables in M as parameters. We then have a map I : ADR (U × M) → ADR (M) which is compatible with the coboundary d and which induces H(I) in cohomology. ∗ (ϕω ) on U × M To any ω ∈ ADR (M) we associate the form L∗ (ω) ∧ πG vol and check easily (see [113, page 150]): ρ(ω) = I(L∗ (ω)). In other words, the following diagram is commutative ADR (M)

L∗

/ ADR (U × M)

SSSS SSSS SSSS SSSS ρ )

L (M)

I

/ ADR (M) 5 k k kkkk k k k kkk ι kkkk

For U, we now choose a contractible neighborhood of e. The identity map on U × M is therefore homotopic to the composition U × M

π

/M

j

/ U × M , where π is the projection and j sends

15

16


x to (e, x). By using I ◦ π ∗ = id and the compatibility of de Rham cohomology with homotopic maps, we get: H(I) ◦ H(L∗ ) = H(I) ◦ idH(M×U) ◦ H(L∗ ) = H(I) ◦ H(π ∗ ) ◦ H(j∗ ) ◦ H(L∗ ) = H(j∗ ) ◦ H(L∗ ) = H((L ◦ j)∗ ) = id . This implies id = H(ι) ◦ H(ρ) and H(ι) is surjective.

1.7

Cohomology of Lie groups

In this section, we give two structure theorems for the cohomology of a Lie group. The first one comes from the existence of left and right G-manifold structures on G and follows from the results of Section 1.6. The second one, called Hopf’s theorem, gives a precise algebra structure for the cohomology. Recall that L (G) (R (G), I (G)) is the set of left invariant (right invariant, bi-invariant) forms on G. Denote by L (G)L=0 the set of left invariant forms whose Lie derivative (see Section A.2) by any left invariant vector field is zero. Theorem 1.30 Let G be a compact connected Lie group with Lie algebra g. Then we have two series of isomorphisms: (1) L (G) ∼ = R (G) ∼ = ∧g∗ ; (2) HL (G) ∼ = HR (G) ∼ = H ∗ (G; R) ∼ = I (G) ∼ = L (G)L=0 . Remark 1.31 There is one point of view that we do not develop here: the translation of the second line in terms of Lie algebras using the isomorphism L (G) ∼ = ∧g∗ . For that, one needs to know the image of the coboundary d and the Lie derivative L through this isomorphism. This theory is well developed in [114]; we give a glimpse of it in Exercise 1.7. Also see Subsection 3.2.1 for the noncompact case of nilpotent Lie groups. We mention also that the existence of an isomorphism between H ∗ (G; R) and I (G) can be extended to the more general situation of symmetric spaces (see Exercise 1.6). In order to prove the theorem, we first need to determine the derivative of the multiplication and the inverse maps. Lemma 1.32 Let G be a Lie group. Denote by µ : G × G → G the multiplication map and by ν : G → G the inverse map. Then we have: Dµ(g,g ) = DLg + DRg

and

−1 Dνg = − DLg ◦ DRg−1 .

1.7


Proof With the identification T(G × G) ∼ = T(G) ⊕ T(G), we write a vector field on G × G as: ((g, g ), (X, X )) = (g, (g , X )) + ((g, X), g ) ∈ g ×Tg (G) ⊕ Tg (G)× g . Therefore, we have: Dµ(g,g ) (X, X ) = DLg (X ) + DRg (X). From this formula and the equality µ(g, ν(g)) = e, we deduce: 0 = Dµ(g,g−1 ) (X, Dνg (X)) = DLg ◦ Dνg (X) + DRg−1 (X) and

−1 Dνg = − DLg ◦ DRg−1 .

Proof of Theorem 1.30 The first series of isomorphisms is clear. It comes from the triviality of the bundle of left (or right) invariant forms on a Lie group G and the identification between Te (G) and g. The first part of (2), HL (G) ∼ = HR (G) ∼ = H ∗ (G; R), is a consequence of Theorem 1.28. Observe now that I (G) is the set of left invariant forms for the left action of the group G × G on G defined by (g1 , g2 )g = g1 · g · g2−1 . Therefore, Theorem 1.28 implies H ∗ (G; R) = HI (G). The next isomorphism, H ∗ (G; R) ∼ = I (G), will follow immediately from the fact that each bi-invariant form on G is closed, implying HI (G) = I (G). To prove this property, let α be a bi-invariant form on G. We compute, using the left and right invariance of α, the inverse image of α by ν:

−1 ◦ DRg−1 (X1 ), . . .)(g −1 ), ν ∗ α(X1 , . . . , Xk )(g) = α(− DLg = (−1)k α(X1 , . . . , Xk )(g). We then have ν ∗ α = (−1)k α. We now use the fact that I (G) = L (G) ∩ R (G) is stable under the coboundary: ν ∗ (dα) = (−1)k+1 dα d(ν ∗ α) = d((−1)k α) = (−1)k dα, which implies dα = 0. The last isomorphism is established in the next proposition. Now recall from Section A.2 that a form α such that L(X)α = 0 for a vector field X is said to be L(X)-invariant.

17

18


Proposition 1.33 Let G be a compact connected Lie group. The L-invariance of a form with respect to left invariant vector fields corresponds precisely to invariance of the form under right translations. In particular, we have: I (G) = ω ∈ L (G) | L(X)ω = 0 for any left invariant vector field X . Proof Let X be a left invariant vector field. We know that X is determined by its value Xe ∈ Te (G) and that (see Definition 1.14) the exponential map exp : Te (G) → G is defined by θt (e) = exp(tXe ), where θ is the 1-parameter subgroup associated to X. There is also an exponential map at any g ∈ G obtained by requiring exp(tXg ) = Lg (exp(tXe )), where Lg denotes left translation by g. Using the left invariance of X (i.e. DLh (Xe ) = Xh for all h ∈ G) we have θt (g) = exp(tXg ) = Lg (exp(tXe )) = g · exp(tX). In conclusion, the flow acts on g by right translation. On the other hand, if θ is the 1-parameter subgroup associated to X, the t Lie derivative satisfies the formula θt∗ ω−ω = 0 θs∗ L(X)ω ds (see Section A.2 and Exercise A.2). Therefore, the form ω is right invariant for the action of elements in the image of the exponential if and only if L(X)ω = 0 for any left invariant vector field X, see [113, Proposition VI, page 126]. Since, in a connected compact group, the exponential is an epimorphism, we get the result. Theorem 1.34 (Hopf’s theorem) If G is a compact connected Lie group, then there exist elements of odd degree, x2pi +1 ∈ H 2pi +1 (G; Q), such that, as an algebra,

H ∗ (G; Q) = ∧ x2p1 +1 , . . . , x2pr +1 . In fact the number of generators is the rank of the group (see Theorem 3.33). Remark 1.35 One can determine the elements x2pi +1 from the structure law of G as follows. Denote by µ∗ : H ∗ (G; Q) → H ∗ (G; Q)⊗H ∗ (G; Q) the map induced by the multiplication µ : G × G → G. An element h ∈ H ∗ (G; Q) is called primitive if µ∗ (h) = 1 ⊗ h + h ⊗ 1. The set PG of primitive elements is a vector space and we have H ∗ (G; Q) ∼ = ∧PG . We will not use this in the sequel and we do not give a proof, instead referring to [113, Theorem IV, page 167], for instance, for an argument with coefficients in R. This approach relies on the general result that a commutative Hopf algebra over a field of characteristic zero is generated, as algebra, by the primitive elements (see [197]). Proof of Theorem 1.34 Let us denote H ∗ (G; Q) by H and consider Q(H) =

H + / H + · H + , the space of indecomposables of the augmentation ideal H + = j>0 H j of the algebra H. We choose a homogeneous basis (xj )1≤j≤r

1.7


of the Q-vector space Q(H). By fixing a section to the canonical surjection H + → Q(H), we consider the elements xj as elements of H. Fact 1: The elements xj are of odd degree. Suppose that some xk is of even degree and let H k be the quotient algebra of H by the ideal generated by the elements xj for j = k. The canonical map qk : H → H k is a morphism of algebras. Denote by µ∗ : H → H ⊗ H the map induced by the multiplication of G and by µ : H → H k ⊗ H k the composition of µ with qk ⊗ qk : H ⊗ H → H k ⊗ H k . Since the algebra H is finite, we know that an integer l exists such that xlk = 0 and xl+1 = 0. Let ρ : G → G × G denote either of the k inclusions g → (g, e) or g → (e, g), where e is the identity element of G. Then clearly the composition obeys µρ = idG . Therefore, we have, for any y ∈ H, µ∗ (y) = 1 ⊗ y + y ⊗ 1 + y with y ∈ H + ⊗ H + . This implies µ(xk ) = xk ⊗ 1 + 1 ⊗ xk . The map µ, being a morphism of algebras, gives µ(xk )

l+1

= (xk ⊗ 1 + 1 ⊗ xk )

l+1

l l+1 xik ⊗ xl+1−i = . k i i=1

On one side, we have µ(xk )l+1 = µ((xk )l+1 ) = 0 and, on the other side, l+1 xik ⊗ thanks to the lack of cross-relations in H k ⊗H k , we get li=1 i = 0. This contradiction implies Fact 1. xl+1−i k By sending the xj to the chosen elements of the basis and extending this correspondence multiplicatively, we define a morphism of algebras φ : ∧ (x1 , . . . , xr ) → H. Fact 2: The morphism φ is an isomorphism. By construction, φ is surjective so we are reduced to establishing its injectivity. Observe that the restriction of φ to ∧ (x1 ) is injective. We

argue by induction and suppose that its restriction to ∧ x1 , . . . , xk−1 is injective. Let a ∈ ∧ (x1 , . . . , xk ) be such that φ(a) = 0. We decompose a into

with a1 and a2 composition ∧ (x1 , . . . , xk )

a = a1 + xk a2 ,

in ∧ x1 , . . . , xk−1 . We denote by φ k the following

φ

/H

µ∗

/H ⊗ H

qk ⊗id

/ H ⊗ H. k

19

20


From φ k (xk ) = 1 ⊗ xk + xk ⊗ 1 and φ k (ai ) = 1 ⊗ ai for i = 1, 2, we deduce 0 = φ k (a) = 1 ⊗ a1 + (xk ⊗ 1 + 1 ⊗ xk )(1 ⊗ a2 ). This implies xk ⊗ a2 = 0 and a2 = 0. From the induction hypothesis, we now get a = a1 = 0 and the restriction of φ to ∧ (x1 , . . . , xk ) is injective as expected. The method used in the previous proof consists essentially in killing elements by taking quotients of algebras by ideals. This is an important technical argument in the theory of minimal models as we will see in the next chapters. As an illustration, we use Theorem 1.30 to compute the first cohomology group of a compact Lie group. Proposition 1.36 Let G be a compact Lie group with associated Lie algebra g. Then H 1 (G; R) ∼ = Hom(Z (g), R), where Z (g) is the center of the Lie algebra g. Proof Let ω ∈ g∗ be a left invariant 1-form. By definition (see Section A.2) we have: dω(X, Y) = Xω(Y) − Yω(X) − ω([X, Y]). Since the form ω is left invariant, this is also true for the functions ω(Y) and ω(X). Since left invariant functions are constant, the previous formula reduces to dω(X, Y) = −ω([X, Y]). Therefore, the form ω is closed if and only if ω takes the value 0 on [g, g] (i.e. ω ∈ [g, g]⊥ ). Now recall the definition of the center of g: Z (g) = X ∈ g | [X, Y] = 0 for any Y ∈ g . Let F be a positive definite invariant symmetric bilinear form on g (which exists by Lemma 1.38). From F(X, [Y, Z]) = F([X, Y], Z), we see that X is in the F-orthogonal complement [g, g]⊥ of [g, g] if and only if X ∈ Z (g). Thus, closed left invariant 1-forms are dual to elements of Z (g). Since HL (G) ∼ = H ∗ (G; R) and left invariant functions are constant, we have the result. Remark 1.37 Observe that, in the previous proof, we have established that [g, g] = g if the center Z (g) of the Lie algebra g is zero. Lemma 1.38 On any Lie algebra g of a compact Lie group G, there is a positive definite symmetric bilinear form F : g × g → R

1.8

Simple and semisimple compact connected Lie groups

such that: 1. F(Ad(g)(X), Ad(g)(Z)) = F(X, Z); 2. F([X, Y], Z) = F(X, [Y, Z]) for any triple (X, Y, Z) of elements of g. Such an F is said to be invariant. Proof Let F be any positive definite bilinear form on g and set F(Ad(g)(X), Ad(g)(Z))dg, F(X, Z) = G

where Ad(g)(X) = ((DRg )−1 ◦ (DLg ))(X). This bilinear form satisfies F(Ad(g −1 )(X), Z) = F(X, Ad(g)(Z)) for any g in G. The result follows now from the fact that the derivative of Ad(g) is the bracket in g (see Proposition 1.11).

1.8


We now come to our first concrete application: the vanishing of the second Betti number of a compact semisimple Lie group and the determination of the third Betti number of a simple Lie group. As we note in Remark 1.50, one can, in fact, do better and prove that the second homotopy group is zero and the third homotopy group of a simple Lie group is Z. We will come back to this point in Section 1.11. Definition 1.39 A connected compact Lie group is simple if it does not contain any nontrivial connected normal subgroups. A Lie algebra is simple if it has no proper ideal. For instance, SU(n) is simple (see Theorem 1.53) while U(n) is not because it contains SU(n) as a normal subgroup. Remark 1.40 Let G be a Lie group with a simple Lie algebra g. From the correspondence between normal Lie subgroups of G and Lie ideals of g, we see that a normal Lie subgroup H = G of G has dimension zero, so it is discrete (remembering that it is closed). With a similar argument for the converse, we have proved that a Lie group is simple if and only if its Lie algebra is simple. There exist several equivalent definitions of semisimple Lie group. Since we are concerned with real cohomology, we use the following. Definition 1.41 A compact connected Lie group is semisimple if its first Betti number is zero. That is, H 1 (G; R) = 0.

21

22


Proposition 1.42 For a connected compact Lie group G, the following conditions are equivalent: 1. 2. 3. 4.

G is semisimple; the fundamental group π1 (G) ∼ = H1 (G; Z) is finite; the center of G is finite; of G is compact. the universal covering group G

Proof The equivalence of (1) and (2) is a direct consequence of the universal coefficient theorem which implies H 1 (G; R) = Hom(H1 (G; Z), R). As for the equivalence between (2) and (4), observe that a covering space is compact if and only if the fiber is finite. The equivalence of (1) and (3) is a consequence of Proposition 1.36. Remark 1.43 Because the Lie algebra associated to the center of G is the center Z (g) of the Lie algebra g, we observe that the center Z (g) is zero if g is the Lie algebra of a connected compact semisimple Lie group. Proposition 1.44 Each simple Lie group is semisimple. Proof Since the group is simple, its center must be finite and the result follows from Proposition 1.42. A product of two simple Lie groups is an example of a semisimple Lie group that is not simple. Theorem 1.45 If G is a compact semisimple Lie group, then the second Betti number b2 (G) is zero. That is, H 2 (G; R) = 0. Corollary 1.46 If G is a compact semisimple Lie group, then π2 (G)⊗Q = 0. Theorem 1.45 is a direct consequence of the definition of semisimple Lie group and of Hopf’s theorem (Theorem 1.34). To emphasize the interrelationship between geometry and homotopy theory, we will give two other proofs of the theorem, one using the material we have just discussed on invariant forms and a second one in Section 1.11, using the existence of a universal bundle for a Lie group (see page 40). Proof 2 of Theorem 1.45 Suppose α ∈ 2I (G). By Theorem 1.30, if we can show that α = 0, then this will imply H 2 (G; R) = 0. Expressing the L-invariance of α for a left invariant vector field X gives: 0 = L(X)α = (i(X)d + di(X))α = i(X)dα + di(X)α = di(X)α, since α is closed. Hence, i(X)α is a closed 1-form. Since G is semisimple, we have H 1 (G; R) = 0, so i(X)α is an exact 1-form. That is, there exists a smooth function f : G → R such that i(X)α = df . By Proposition 1.26,

1.8


we know that i(X)α is left invariant. By Theorem 1.28, left invariant cohomology is isomorphic to ordinary cohomology, so i(X)α must be exact by a left invariant function as well. Therefore (see Remark 1.8), f is constant and i(X)α = df = 0 for any left invariant vector field X. We show now that this implies α = 0. For this, let g ∈ G and choose any two vectors vg , wg ∈ Tg (G). We can find left invariant vector fields V and W with Vg = vg and Wg = wg simply by left translating vg , wg around G. But because V is left invariant, we then have αg (vg , wg ) = αg (Vg , Wg ) = α(V, W)(g) = i(V)α(W)(g) = 0 since i(X)α = 0 for all left invariant X. Since g, vg and wg were arbitrary, we have α = 0. Therefore, no nonzero invariant 2-forms exist on G and H 2 (G; R) = 0. Proof of Corollary 1.46 Since a compact semisimple Lie group is finitely covered by a compact simply connected semisimple Lie group, it suffices to prove the result in the simply connected case. The Hurewicz theorem then implies that π2 (G) ∼ = H2 (G), so from H 2 (G; R) = Hom(H2 (G), R) = 0, we deduce that π2 (G) is finite. Theorem 1.47 If G is a compact semisimple Lie group, then the third Betti number b3 (G) is greater than or equal to one. In the simple case, we have H 3 (G; R) = R. Observe that, as a direct consequence, we have the following. Corollary 1.48 The only spheres which have a Lie group structure are S0 , S1 , S3 . With the same argument as in the proof of Corollary 1.46, we also have the following. Corollary 1.49 If G is a compact semisimple Lie group, then π3 (G)⊗Q = 0. In the simple case, we have π3 (G) ⊗ Q = Q. Remark 1.50 Since we are interested only in the rational or real world, we are satisfied with the results of Corollaries 1.46 and 1.49. In fact, however, it is possible to prove that π2 (G) = 0, in the semisimple compact case, and π3 (G) = Z in the simple case. The proofs need material that we do not introduce here: • for π2 (G) = 0, see [32] or [50]; • for π3 (G) = Z in the simple compact case, see [36] or [199, Theorem

4.17, page 335].

23

24


We will come back to the case of π2 (G) = 0 in Remark 1.82 with a more homotopical argument. Proof of Theorem 1.47 Recall first, from Theorem 1.30, that H 3 (G; R) is isomorphic to the vector space of invariant 3-forms on G. We follow the proofs of [107, Problem IV-B] and [41, Section V-12]. The idea is to transform the problem of finding an invariant 3-form on G into the problem of finding an invariant symmetric bilinear form on g. Because such a form exists by Lemma 1.38, we will get the first part of the statement. Let B (g) be the set of symmetric bilinear forms on g such that F([Z, X], Y) = F(X, [Z, Y]) for any triple (X, Y, Z) of elements of g. For such an F, we define ωF by ωF (X, Y, Z) = F([X, Y], Z). We check easily that ωF ∈ ∧3 g∗ ∼ = 3L (G). We prove now that ωF ∈ 3I (G). Let g ∈ G. We have, by invariance of F and the definition of ωF : ωF (Ad(g)(X), Ad(g)(Y), Ad(g)(Z)) = F([Ad(g)(X), Ad(g)(Y)], Ad(g)(Z)) = F(Ad(g)([X, Y]), Ad(g)(Z)) = ωF (X, Y, Z). ∼ Therefore, we have constructed a linear map χ from B (g) to 3I (G) = 3 H (G; R). Since the group G is semisimple, we know (see Remark 1.37 and Remark 1.43) that [g, g] = g, which gives the injectivity of χ . We now have to prove that χ is onto. For that, let ω ∈ 3I (G) and let X ∈ g. From L(X) = di(X) + i(X)d, L(X)ω = 0 and dω = 0, we deduce di(X)ω = 0 with i(X)ω ∈ A2DR (G). By Theorem 1.45, this implies the existence of a 1-form αX such that i(X)ω = dαX . We define F : g × g → R by F(X, Y) = αX (Y). From the definition of the coboundary d and the construction of F, we deduce: F(X, [Y, Z]) = αX ([Y, Z]) = dαX (Y, Z) = i(X)ω(Y, Z) = ω(X, Y, Z). The invariance of ω and the fact that [g, g] = g implies the invariance of F; that is, F(Ad(g)(X), Ad(g)(Y)) = F(X, Y). We are reduced to proving the symmetry of F. For that, observe that F can be uniquely decomposed in F1 + F2 with F1 symmetric and F2 skew. The F2 gives an invariant 2-form on G which must be zero by Theorem 1.45 and therefore F is symmetric. From this first part, we deduce an isomorphism B (g) ∼ = 3I (G). Since B (g) contains a nontrivial element, we have b3 (G) ≥ 1. We suppose now that G is simple and we prove b3 (G) = 1. Let F be a positive definite invariant bilinear form on g and let F ∈ B (g). We

1.8


consider the least value λ of the F (X, X) when X is such that F(X, X) = 1. Set F (X, X) = λ F(X, X) − F (X, X). The kernel of F is a subspace of g, invariant under the bracket (since F is invariant). This kernel is not equal to 0 because λ is reached by F , so it must be equal to all of g since g has no proper ideal. We get F = λF, which means that the dimension of B (g) is 1. Remark 1.51 Finally, observe that, in the case of a semisimple Lie group G, the negative of the Killing form (X, Y) → −trace (ad(X) ◦ ad(Y)) is a nondegenerate bilinear symmetric form on the Lie algebra g. If G is compact, then it can be shown that this symmetric bilinear form is positive definite as well. Since G is parallelizable, this then defines a metric on the tangent bundle. With respect to this metric we can define a Hodge star operator and obtain a Hodge decomposition of forms on G (see Section A.4). From [112] and [107], we can see that if G is a compact semisimple Lie group, then the harmonic forms are the invariant forms. So, together with Proposition 1.33, the harmonic forms are the left invariant forms which are also L-invariant. Because, L(X) is a derivation, this implies that, on a compact semisimple Lie group, the wedge product of harmonic forms is harmonic. This is an unusual property that can be extended to symmetric spaces (and, in the language of Chapter 2, imbues them with the property of formality) (see Exercise 3.8). We do not go further in this direction, instead referring the reader to [107, Propositions 4.4.2 and 4.4.3] for more details in the semisimple Lie group case. We end this section with Cartan’s theorem on the classification of simple Lie groups. Since this lies outside the main subject of this book, we do not give proofs, and, rather, refer the reader to [2], [112], [199], [214], [229] and [262] for presentations of this theory. Cartan’s theorem gives a decomposition of compact connected Lie groups into a product of particular Lie groups and gives a complete classification of these factors. For this, semisimple Lie algebras are the key notion because semisimple Lie algebras allow a global Jordan decomposition for any representation (see [103, page 129]). Since the Lie algebra g of a Lie group G can be decomposed into

g=Z⊕

k

gi ,

i=1

where Z is the center of g and gi are simple ideals, the following can be shown (see [199, page 282]).

25

26


Theorem 1.52 For an arbitrary compact connected Lie group G, there exists a torus T, compact simple simply connected Lie groups G1 , . . ., Gk and a finite group K contained in the center of T × G1 × · · · × Gk such that the quotient (T × G1 × · · · × Gk ) /K is isomorphic to G. In other words, any compact connected Lie group admits a finite sheeted covering group which is the product of a torus and of simple Lie groups. For instance, any connected compact Lie group with a trivial center is isomorphic to a product of simple compact Lie groups. This is the beginning of the root system construction. The second part is the classification of compact connected simple Lie groups. Theorem 1.53 The following groups are simple: 1. 2. 3. 4.

the special unitary groups SU(n1 ); the special orthogonal groups SO(n2 ); the symplectic groups Sp(n3 ); the exceptional Lie groups G2 , E6 , E7 , E8 , F4 ;

and any compact connected simple Lie group is isomorphic to one group of this list. If we insist that n1 ≥ 2, n2 ≥ 5, n3 ≥ 2, there are no isomorphisms between two elements of this list.

1.9

Homogeneous spaces

We present the definition of locally trivial fiber bundles here and an important tool for their construction, the pullback along a map with values in the base. Homogeneous spaces are then introduced and studied as examples of bundles. General references are [112], [145] and [242]. Definition 1.54 A locally trivial fiber bundle of fiber F is a continuous map p : E → B together with a space F such that B admits a numerable open cover (Ui ) with homeomorphisms ϕi making commutative the following diagram Ui × F EE EE EE EE p1 E"

ϕi ∼ =

Ui

/ p−1 (U ) i xx x xx xx p x {x

Here p1 is the canonical projection. The space B is called the base and E is called the total space. The collection (Ui ) is called a trivializing open cover of the base and the homeomorphisms ϕi the local trivializations. If all spaces

1.9

Homogeneous spaces

are manifolds and all maps are smooth maps, we say that we have a smooth locally trivial fiber bundle. Let p : E → B and p : E → B be two locally trivial fiber bundles. A morphism between p and p is a pair of maps, (ψ, ), such that the following diagram commutes E p

B

/ E

ψ

p

/ B

If ψ is the identity on B and a homeomorphism, we say that p and p are in the same isomorphism class of bundles. Example 1.55 • For any pair of topological spaces, the canonical projection B × F → B

is a locally trivial fiber bundle. We call it the trivial bundle. • If the base B of the bundle E → B is paracompact Hausdorff, then any

open cover of B is numerable, so the restriction in the definition is really not important for us throughout the book. Example 1.56 (Pullback of a locally trivial fiber bundle) If p : E → B is a locally trivial fiber bundle and ψ : B → B is a continuous map, then we define a topological space ψ ∗ E = (b, x ) ∈ B × E | ψ(b) = p (x ) = b × p −1 (ψ(b)) b∈B

as a subspace of the product B × E . The two projections ψ ∗ E

j

/ E ,

p

/ B, (b, x ) → b, are obviously continuous. By (b, x ) → x , and ψ ∗ E construction, for any commutative diagram

E p

B

/ E

ψ

p

/ B

there exists a unique continuous map : E → ψ ∗ E such that j ◦ = and p ◦ = p. This map is defined by (y) = ( p(y), (y)). Therefore,

27

28


our construction ψ ∗ E satisfies a universal property. Moreover, if the map p : E → B is the canonical projection B × F → B , then we have ψ ∗ E = B × F and the map p : ψ ∗ E → B is the trivial bundle on B. We show now that p : ψ ∗ E → B is a locally trivial fiber bundle with fiber F. Since p is locally trivial, we may choose a trivializing open cover (Ui ) of B together with homeomorphisms ϕi such that the following triangle commutes Ui × F DD DD DD DD D!

ϕi ∼ =

Ui

/ p −1 (U ) i x x x xx xx p x |x

Set Ui = ψ −1 (Ui ) and consider the diagram, / E < z zz z z zz zz

ψ ∗ E

{= {{ { {{ {{

Ei

Ui

B {= {{ { { {{ {{

ϕi

/ U × F i ψ

/ U i

p

/ B < yy yy y yy yy

where Ei is the pullback construction applied to the trivial fibration Ui × F → Ui and the map Ui → Ui induced by ψ. We know, from the remark above, that the map Ei → Ui can be identified with the canonical projection Ui × F → Ui . We deduce that the (Ui ) define a trivializing open cover of B and that p : ψ ∗ E → B is a locally trivial fiber bundle, called the pullback of p : E → B by the map ψ. In the case of a smooth locally trivial fiber bundle, one has to define a manifold structure on ψ ∗ E such that the various maps are smooth. This is a standard procedure (see [112, Proposition VII, page 29]). Below, we look at certain geometrically important examples of locally trivial fiber bundles. We shall freely use various properties of bundles such

1.9

Homogeneous spaces

as the homotopy lifting property and the existence of the long exact homotopy sequence. See Exercises 1.9 and 1.10. In the context of Lie groups, the main examples of locally trivial fiber bundles come from the notion of homogeneous spaces that we introduce now. Let H be a closed subgroup of a Lie group G. We denote by G/H the set of left cosets of H; that is, G/H is the quotient of G by the equivalence relation x ∼ y if and only if x−1 y ∈ H. The elements of G/H are denoted by xH for x ∈ G. In particular, H is the class of the neutral element e of G. We denote by q : G → G/H the canonical projection. Definition 1.57 The space G/H constructed above is called a homogeneous space. Definition 1.58 Let H be a closed subgroup of G. A local section of H in G is a continuous map σ : U → G, defined on an open neighborhood U of H ∈ G/H such that q ◦ σ = idU . Proposition 1.59 If H is a closed subgroup of G admitting a local section, then, for any closed subgroup K of H, the canonical projection p : G/K → G/H, gK → gH is a locally trivial bundle of fiber H/K. O(n) has a local section. O(n − k) Consider (e1 , . . . , en ) ∈ O(n); we define an open set U of the quotient as O(n) n (e1 , . . . , en−k , v1 , . . . , vk ) is a basis of R . U = (v1 , . . . , vk ) ∈ O(n − k) Remark 1.60 The canonical map O(n) →

From the Gram–Schmidt orthonormalization procedure, if (u1 , . . . , un ) is a basis, an orthonormal basis is constructed by GS(u1 , . . . , un ) = (u 1 , . . . , u n ) with u n =

un , un

u n−1 =

un−1 − un−1 , u n u n , un−1 − un−1 , u n u n

... .

Therefore, we obtain a section σ : U → O(n) of the canonical projection defined by σ (v1 , . . . , vk ) = GS(e1 , . . . , en−k , v1 , . . . , vk ) = (e1 , . . . , , v1 , . . . , vk ). en−k More generally, in the case of a closed subgroup of a Lie group, there always exists a local section (see [60, 12, Proposition 1]). We do not give

29

30


the proof here. In every concrete example, a local section can be easily constructed as above. Proof of Proposition 1.59 Observe first that any local section σ : U → G, defined on a neighborhood U of H in G/H, gives a local section σg : gU → G, σ (g H) = g.σ (g −1 g H), with domain the neighborhood gU of gH. Let x ∈ G/H with a local section (U, σ ). To satisfy the requirements of the definition of a locally trivial bundle, one needs maps ϕ and ψ such that ϕ ◦ ψ = idp−1 (U) , ψ ◦ ϕ = idU×G/K and the following diagram commutes ψ

r

U × (G/K) NNN NNN NNN NN&

ϕ

/ p−1 (U) ⊂ G/K oo ooo o o oo w oo p o

U ⊂ G/H

We define ϕ(y, hK) = σ (y)hK, ψ(gK) = (gH, (σ (gH))−1 gK) and check easily that ϕ(ψ(gK)) = ϕ(gH, (σ (gH))−1 gK) = σ (gH).(σ (gH))−1 gK = gK, ψ(ϕ(y, hK)) = ψ(σ (y)hK) = (σ (y)hH, (σ (σ (y)H))−1 σ (y)hK), = (pσ (y), (σ (p(σ (y)))−1 σ (y)hK) = (y, hK), pϕ(y, hK) = p(σ (y)hK) = σ (y)hH = σ (y)H = pσ (y) = y.

Our goal later will be to create algebraic models for manifolds and Lie groups and homogeneous spaces in particular. But the algebraic models we will consider work best in the simply connected world – or at least in the nilpotent world (see Definition 2.32 and the discussion that follows). For Lie groups and homogeneous spaces, this will not present a problem, for we prove at the end of this section that Lie groups and homogeneous spaces are simple, and therefore nilpotent, spaces. Definition 1.61 A space X is said to be simple if its fundamental group is abelian and acts trivially on the higher homotopy groups of X. For completeness, we recall the definition of the action of the fundamental group on the other homotopy groups. Let α ∈ π1 (X), ξ ∈ πn (X). The inclusion of the basepoint s0 → Sn is a cofibration (see [265] or [240]), so

1.9

Homogeneous spaces

there exists a commutative diagram as follows s0 × 0

/ s0 × I

α

Sn

×0

/ Sn × I

ξ

PP P PF PP P' 0 X

Now, the action of α on ξ , denoted α · ξ is defined to be α · ξ = F(−, 1), and its homotopy class does not depend on the choice of F. Proposition 1.62 Suppose H is a connected closed subgroup of a compact connected Lie group G. Then the homogeneous space G/H is a simple space. In particular, G itself is a simple space. In fact, this result holds for any topological group or, more generally, for any so-called H-space. Proof Denote the quotient map by q : G → G/H. We use the trivial class H as base point for G/H. Let α ∈ π1 (G/H), ξ ∈ πn (G/H). Since H is connected, there is a surjection q# : π1 (G) → π1 (G/H), so choose some α : I → G with α (0) = α (1) = e and q( α (t)) = α(t). (The equality can be achieved since q satisfies the homotopy lifting property, see Exercise 1.9.) Now, by using the left action of G on G/H, we define a map F making the diagram commute: F(x, t) = α (t) ξ(x). Observe F(x, 0) = ξ(x), F(x, 1) = ξ(x) and α (t) ξ(s0 ) = α (t) H = α(t). F(s0 , t) = The last equality shows that F makes the diagram commute. Moreover, we see that F(−, 1) = ξ . Hence, α · ξ = ξ and the action is trivial. Example 1.63 Consider the following examples: 1. Tori T n have π1 (T n ) = Zn , but are simple spaces. 2. The special orthogonal groups SO(n) have π1 (SO(n)) = Z/2, but are simple spaces. 3. The projective space RP(2n) is known not to be simple (i.e. the antipodal map on the universal covering S2n has degree −1), but we have RP(2n) = O(2n + 1)/(O(2n) × O(2)). Note that the subgroup H is not connected.

31

32


Remark 1.64 The special case H = T, where T is a maximal torus of G, is easier to handle because there is a Bruhat decomposition of G/T showing that a CW-structure for G/T has cells only in even dimensions. Hence, the homogeneous space G/T is simply connected. A proof using regular elements can be found in [199, page 277]. Also, compare Exercise 1.5.

1.10

Principal bundles

In this section, we define principal bundles with structure group a Lie group G (also called principal G-bundles). Since the pullback of a principal Gbundle is a principal G-bundle, we look for a principal G-bundle p : E → B which is universal in the following sense: any principal G-bundle over a space B can be obtained as a pullback of p along a map B → B. A characterization of such bundles is proved by using the notion of CWcomplexes. Other examples will be developed in Section 1.12. Definition 1.65 Let G be a Lie group. A principal bundle with structure group G (or principal G-bundle) is a locally trivial bundle p : E → B with fiber the Lie group G, together with a right action E × G → E, (x, g) → xg, of G on E and a trivializing open cover of B, (Ui , ϕi ), such that ϕi (x, g ·g ) = ϕi (x, g)g , for any x ∈ E, g, g ∈ G. If all spaces are manifolds and all maps are smooth maps, we call the principal bundle a smooth principal bundle with structure group G (or smooth principal G-bundle). Observe that the previous definition makes sense for a topological group instead of a Lie group G. Remark 1.66 As the reader can easily check, if p : E → B is a principal G-bundle then the right action of G on E is free. Reciprocally, if G is a Lie group that acts freely and properly on a manifold M, the canonical projection M → M/G is a principal G-bundle, see [113, pages 193 and 229]. Example 1.67 (1) For any space B, the canonical projection B × G → B is a principal G-bundle, called the trivial principal G-bundle. (2) If H is a closed subgroup of a Lie group G, the canonical map G → G/H is a principal H-bundle. Definition 1.68 A morphism between two principal G-bundles is a pair of maps, (ψ, ), such that is compatible with the G-action (i.e. (xg) =

1.10

Principal bundles

(x)g) and the following diagram commutes E p

/ E

B

p

/ B

ψ

Definition 1.69 A principal G-bundle p : E → B is trivial if there exists a morphism of principal bundles B×G p

f

/ E

B

B

p

where B × G → B is the canonical projection. Observe that, together with Exercise 1.11, if a principal G-bundle is trivial, the map f of Definition 1.69 is a homeomorphism. Proposition 1.70 Let G be a Lie group and let p : E → B be a principal G-bundle. If p admits a section, then p is trivial. Proof Denote by σ the section of p and let B × G → B be the canonical projection. We construct a morphism of principal G-bundles B×G

B

/ E p

B

by (x, g) = σ (x) · g. The triviality follows by definition.

Example 1.71 (Pullback of a principal G-bundle) Let G be a Lie group and p : E → B be a principal G-bundle. Let ψ : B → B be a continuous map. Recall from Example 1.56 the construction of the pullback of p, denoted p : ψ ∗ E → B . This is a locally trivial fiber bundle with fiber G. We define a free right action of G on ψ ∗ E by (x, g)h = (x, g · h). This gives ψ ∗ E the structure of principal G-bundle and we call p : ψ ∗ E → B the pullback principal G-bundle of p : E → B along the map ψ. The same construction gives a smooth principal G-bundle if p : E → B and ψ : B → B are smooth.

33

34


Observe, from Exercise 1.11, that for any morphism of principal Gbundles, E p

B

/ E

ψ

p

/ B

the domain E is in the isomorphism class of ψ ∗ E. Let p : E → B be a fixed principal G-bundle. Using Example 1.71, to any map f : X → B we can associate a principal G-bundle over X. In fact, if we consider the isomorphism classes of principal G-bundles, this association depends only on the homotopy classes of maps (see Exercise 1.8). Denote by kG (X) the set of isomorphism classes of principal G-bundles over X. What we have described above is a correspondence [X, B] → kG (X), [f ] → f ∗ E. We can ask whether there exist principal G-bundles p : E → B for which this association is an isomorphism (compare with [145], [74, Classification Theorem], [242]). This will elicit the notion of classifying space BG of G developed in Section 1.11. Because classifying spaces are not naturally manifolds (although, by [77], they can be viewed as manifolds of infinite dimension), we must leave the framework of manifolds and enter that of CW-complexes. This is still convenient for our study because any compact manifold is a CW-complex, as classical Morse theory shows (see [195]), and also because two simply connected compact Lie groups are isomorphic if they are homotopy equivalent (see [231]). Definition 1.72 Let G be a Lie group. A principal G-bundle p : E → B is an n-universal G-bundle if the association given above, [X, B] → kG (X), is an isomorphism for any CW-complex X of dimension ≤ n. A universal principal G-bundle is an n-universal principal G-bundle for every n. Universal principal G-bundles can be characterized by the following result. Theorem 1.73 A principal G-bundle p0 : E0 → B0 is n-universal if and only if the space E0 is (n − 1)-connected. This criterion will give us the existence of universal G-bundles for the classical examples of Lie groups (see Section 1.12). For instance, in the

1.10

Principal bundles

case of G = S1 or G = S3 we can explicitly find universal bundles as follows. Example 1.74 Consider the classical action of the circle S1 on S2n+1 ⊂ Cn+1 given by the complex multiplication (z1 , . . . , zn+1 )z = (z1 · z, . . . , zn+1 · z). The quotient is the complex projective space and we have a principal S1 bundle, S2n+1 → CP(n), which is (2n + 1)-universal by Theorem 1.73. The canonical inclusions CP(1) ⊂ CP(2) ⊂ · · · ⊂ CP(n) ⊂ · · · and S1 ⊂ S2 ⊂ · · · ⊂ Sn ⊂ · · · define spaces CP(∞) = ∪n CP(n) and S∞ = ∪n Sn . The principal S1 -bundle S∞ → CP(∞) is a universal S1 -bundle since S∞ is contractible. By using the quaternionic multiplication on S3 , we get a (4n+3)-universal principal S3 -bundle, S4n+3 → HP(n), and a universal principal S3 -bundle, S∞ → HP(∞). Proof of Theorem 1.73 Part 1. Suppose that p0 : E0 → B0 is n-universal and let f : Sk → E0 be a representative of a homotopy class in πk (E0 ), k ≤ n − 1. We consider the following morphism of principal G-bundles (which is therefore a pullback), Sk × G

Sk

/ E0

ψ

p0

/ B0

defined by ψ = p0 ◦ f and (x, g) = f (x)g. The pullback of p0 along ψ being trivial, the map ψ must be homotopically trivial by the injectivity part of the hypothesis and there is an extension ψ : Dk+1 → B0 of ψ. From the surjectivity part of the hypothesis, we now get the following morphism of principal G-bundles

Dk+1 × G

Dk+1

ψ

/ E0

p0

/ B0

Using the injectivity part again gives the following sequence of morphisms of principal G-bundles where the maps Sk → Dk+1 and Sk ×G → Dk+1 ×G

35

36


are the canonical injections: Sk × G

/ Dk+1 × G

/ Dk+1

Sk

/ E0

ψ

p0

/ B0

The restriction of to Dk+1 × {e} is an extension of f : Sk → E which implies that f is nullhomotopic. Part 2. Suppose now that p0 : E0 → B0 is a principal G-bundle such that πk (E0 ) = 0 for k ≤ n − 1. We proceed by induction on the dimension of the CW-complex X, the statement obviously being true if X is of dimension 0. We suppose the result to be true for any CW-complex L of dimension less than or equal to k − 1 with k ≤ n. We attach a k-cell to L by a map χ : Sk−1

χ

/L

j

/ X = L ∪ Dk .

For proving the surjectivity part of the statement, we consider a principal Gbundle p : E → X. The pullback (j ◦χ)∗ E being trivial, we have a morphism (χ , χ ) from the trivial bundle to j∗ E. The induction hypothesis (surjective part) applied to j∗ E → L gives a morphism (ψL , L ) of principal G-bundles as in the following diagram χ

Sk−1 × G KKK KKK KKK K% Dk × G

Sk−1

χ

LLL LLL LLL LL %

Dk

/ j∗ E

L

@@ @@ @@ @@ / E

ψL / L @@ p @@ @@ @@ j / X

/ E0 ?

p0

/ B0 ? ψ

Since the map L ◦χ(−, e) : Sk−1 → E0 is trivial by assumption, there exists : Dk × G → an extension : Dk → E0 of it. We now extend to a map E0 using the action of G (i.e. (x, g) = (x)g). Since the total space E is the pushout of j∗ E and Dk × G over Sk−1 × G, we construct a morphism

1.10

Principal bundles

(, ψ)from p to p0 as follows. • if x ∈ p−1 (y), y ∈ L, then (x) = L (x), ψ(y) = ψL (y); −1 (x)), ψ(y) = p0 ((y)). • if x ∈ p−1 (y), y ∈ Dk , then (x) = (χ

Now, Exercise 1.11 implies that ψ ∗ E0 and E are in the same isomorphism class. For proving the injectivity part of the statement we consider two maps ψ1 , ψ2 : X → B0 giving two isomorphic bundles ψ1∗ E0 , ψ2∗ E0 . By induction, the maps ψ1 ◦ j and ψ2 ◦ j are homotopic. The homotopy lifting property gives a map FL : L × [0, 1] → E0 such that p0 ◦ FL is the previous homotopy between ψ1 ◦ j and ψ2 ◦ j. Sk−1 × [0, 1]

Dk

χ×id

/ L × [0, 1] PPP PPP FL PPP PPP PPP ( F / E0 X × [0, 1] 6 5 k k k k k kkk p0 kkk F kkk

× [0, 1]

B0

From πk−1 (E0 ) = 0, we deduce the homotopy triviality of the composition FL ◦ (χ × id) and, therefore, the existence of an extension F : Dk × [0, 1] → E0 . By the construction of X as a pushout, we get a map F : X × [0, 1] → E0 making the previous diagram commutative. The composition p0 ◦ F is a homotopy between ψ1 and ψ2 . Remark 1.75 John Milnor constructed a universal bundle for any topological group. In short, the construction goes like this: • the total space EG is the infinite join, EG = G ∗ G ∗ G ∗ · · · ; • G acts on EG diagonally by (g1 , g2 , . . .)g = (g1 ·g, g2 ·g, . . .). By definition,

the space BG is the quotient EG/G. Milnor proves that EG → BG is a universal G-bundle. For more details, see [145, Chapter 4, Section 11] or [193]. Here, we do not need this generality. In Section 1.12, we explicitly construct the classifying spaces for G = O(n), G = U(n) and G = Sp(n). Since any compact Lie group is isomorphic to a subgroup of the orthogonal group, the existence of a classifying space for O(n) implies the existence of a classifying space for any compact Lie group. Remark 1.76 In terms of G-spaces and in the case n = ∞, Theorem 1.73 can also be proved by establishing the following property (see [74, 7.7]): If

37

38


E is a free G-space and E a contractible G-space, then there exists a G-map E → E and any two such maps are G-homotopic.

1.11

Classifying spaces of Lie groups

Here we give the definition of the classifying space of a compact Lie group and determine its cohomology algebra with rational coefficients. We use this classifying space to give a new, more homotopical, proof of the fact that the second Betti number of a semisimple Lie group is always equal to 0 (see Theorem 1.45). Definition 1.77 If G → EG → BG is a universal principal G-bundle, the space BG is called the classifying space of the Lie group G. Note that, because EG is contractible, the long exact homotopy sequence of the bundle gives πi (G) ∼ = πi+1 (BG). In fact, (BG) G, where (BG) here denotes the loop space of BG. Remark 1.78 From Remark 1.75, one knows that such spaces exist. The uniqueness of their homotopy type is guaranteed by the following result. Two compact Lie groups G and H are isomorphic if and only if the classifying spaces BG and BH are homotopy equivalent (see [204], [210, Section 6], [211], [218]). For concrete constructions of classifying spaces, the following observation will be of use. Example 1.79 Let G be a compact connected Lie group and H be a closed subgroup of G. Denote by G → EG → BG a universal principal G-bundle. The subgroup H acts freely on EG and gives a principal H-bundle H → EG → EG/H. Since the homotopy groups of EG are trivial, this fibration is a universal principal H-bundle. That means we can take EG/H to be the classifying space BH. Proposition 1.80 G/H → BH → BG is a fibration. Proof As in Proposition 1.59, we have a fibration G/H → EG/H → EG/G. The result follows now with BH = EG/H and BG = EG/G. Theorem 1.81 Let G be a compact connected Lie group with cohomology algebra an exterior algebra, H ∗ (G; Q) = ∧(u1 , . . . , ur ), where ui ∈ H 2ni −1 (G; Q). Then the classifying space for G has for cohomology algebra a polynomial algebra, H ∗ (BG; Q) = Q[v1 , . . . , vr ] with vi ∈ H 2ni (BG; Q). Recall from Hopf’s theorem (Theorem 1.34) that the cohomology algebra H ∗ (G; Q) is always an exterior algebra on odd degree generators for any compact connected Lie group G. We present two different proofs for

1.11


Theorem 1.81. The first one, due to Borel [33], consists of a careful analysis of a spectral sequence. In the second one, we keep track of spaces more carefully. As the reader will see, this process simplifies the technical part of the argument. A third approach, using the construction of the minimal model of a loop space X from the minimal model of X, will be presented in Example 2.67. Proof 1 of Theorem 1.81 Consider the Serre spectral sequence, with coefficients in Q, of the principal bundle G → EG → BG. Note that the hypotheses of the theorem guarantee that BG is simply connected. Therefore, the Serre spectral sequence has simple coefficients and the second page p,q p,0 0,q has the form E2 = E2 ⊗E2 = (∧(u1 , . . . , ur ))p ⊗H q (BG; Q) which satisfies the hypothesis of the Zeeman–Moore theorem (see Theorem B.15). We know from Theorem 1.73 that the page E∞ is reduced to Q in degree 0. Suppose, for a moment, that the ui satisfy the following property: (T )

dj ui = 0 for j ≤ 2ni − 1 and d2ni (ui ) = vi = 0,

where the dj are the differentials in the spectral sequence. Now define a cochain algebra ∧(u1 , . . . , ur )⊗ Q[v1 , . . . , vr ] with dui = vi , |ui | = 2ni − 1, |vi | = 2ni . Observe that the sub-differential algebras (∧(ui ) ⊗ Q[vi ], d) are acyclic. Therefore the cohomology of ∧(u1 , . . . , ur ) ⊗ Q[v1 , . . . , vr ] is reduced to Q in degree 0. We filter by the degree in p,q p,q the vi ’s and get a spectral sequence Er whose second page is E2 = (∧(u1 , . . . , ur ))p ⊗ (Q[v1 , . . . , vr ])q . The canonical map sending ui to ui and vi to vi gives a morphism of specp,q p,q tral sequences (En , d n ) → (En , dn ). This morphism satisfies conditions (1) and (3) of Theorem B.15. Therefore, condition (2) is satisfied also and H ∗ (BG; Q) ∼ = Q[v1 , . . . , vr ]. Elements which satisfy property T are called transgressive and the proof of the fact that the ui ’s are transgressive is one important point in the proof of Borel [33]. We do not go further in this direction. In the second proof we deal with spaces and do not work only at the level of spectral sequences. Recall that each space X has an associated path fibration p

X → PX → X, where PX consists of all continuous paths γ : I → X having γ (0) = x0 , for a fixed basepoint x0 ∈ X, and p(γ ) = γ (1). Then the fiber p−1 (x0 ) consists of all paths with γ (0) = x0 = γ (1). This is the loop space of X, X. It is easy to see that PX is contractible, so the long exact homotopy sequence of the fibrations (see Exercise 1.10) gives πi (X) ∼ = πi+1 (X). The following proof applies the Serre spectral sequence to the path fibration. Proof 2 of Theorem 1.81 We prove the next property by induction on r: Let X be a connected simple space whose loop space X has rational

39

40


cohomology algebra an exterior algebra H ∗ (X; Q) = ∧(u1 , . . . , ur ) with ui of odd degree 2ni − 1. Then H ∗ (X; Q) = Q[v1 , . . . , vr ] is a polynomial algebra with |vi | = |ui | + 1. Since the space X is simple (see Proposition 1.62), it admits a rationalization (see Subsection 2.6.1) and, since we are interested only in rational cohomology, we may replace X by its rationalization. However, for convenience, we still denote the rationalization by X in this proof. If r = 1, we take the argument of the first proof above. Obviously the element u1 must satisfy the property T and we get the result. Suppose now that the result is true for r − 1 generators and let X be as in the statement. If u1 is an element of lowest positive degree among the ui ’s, it must be a transgressive element, and we denote by v1 ∈ H 2n1 (X; Q) the image of u1 by d2n1 . This class v1 corresponds to a map ϕ(v1 ) : X → K(Q, 2n1 ). Denote by Y the homotopy fiber of ϕ(v1 ) and take the following Puppe fibration sequence (see [137] or [265], for instance) Y

/ X

ϕ(v1 )

/ K(Q, 2n1 − 1)

/Y

/X

ϕ(v1 )

/ K(Q, 2n1 ).

/ X / K(Q, 2n1 − 1) is trivial because the Now, the fibration Y map ϕ(v1 ) admits a section (see Exercise 1.12). We thus get H ∗ (X; Q) = ∧(u1 ) ⊗ H ∗ (Y; Q). By quotienting out the ideal generated by ∧(u1 ), we deduce that H ∗ (Y; Q) = ∧(u2 , . . . , ur ). We apply the induction hypothesis to Y and obtain H ∗ (Y; Q) = Q[v2 , . . . , vr ]. Consider the Serre spectral sequence of the fibration /X / K(Q, 2n1 ). Its second page is E2 = Q[v1 ] ⊗ Q[v2 , . . . , vr ]. Y Being totally concentrated in even degrees, all the differentials must be zero and we see that H ∗ (X; Q) = Q[v1 ] ⊗ Q[v2 , . . . , vr ] = Q[v1 , v2 , . . . , vr ].

We now use the existence of the classifying space of a Lie group for a third proof of Theorem 1.45. Let G be a compact semisimple Lie group. Recall that, in Corollary 1.46, we proved π2 (G) ⊗ Q = 0. As we will see in Remark 1.82, this third proof, which is more homotopical in spirit, in fact points the way toward the more general result that π2 (G) = 0. Proof 3 of Theorem 1.45 Since a compact semisimple Lie group is finitely covered by a compact simply connected Lie group, it suffices to prove the result in the simply connected case. The Hurewicz theorem then implies that π2 (G) ∼ = H2 (G), so in order to show H 2 (G; R) = Hom(H2 (G), R) = 0, we need only to show that π2 (G) is finite. Now, π2 (G) ∼ = π3 (BG), where BG is the classifying space of G. Because π2 (G) = H2 (G) is finitely generated, we have a splitting π2 (G) = F ⊕ T, where F is free abelian and T is the torsion subgroup. Let f ∈ F be a generator of the free abelian part of π2 (G) considered as an element in

1.11


π3 (BG) and use f : S3 → BG to induce a principal G-bundle P → S3 by pulling back the universal one. There is an associated Puppe sequence ∂

f

. . . → S3 → G → P → S3 → BG which is exact on homotopy groups. Furthermore, the connecting map ∂ is simply the loop of the classifying map f up to homotopy: f ∂. We have the following commutative diagram:

Z = π3 (S3 ) ∼ =

Z = π2 (S3 )

f#

∂#

/ π3 (BG)

∼ =

/ π2 (G)

from which it follows that ∂# (x) = f , where x generates Z = π2 (S3 ) = H2 (S3 ). Because f represents a generator of F in π2 (G) = H2 (G), we can define a homomorphism φ : H2 (G) → Z as follows: Write the generators of F as {f = f1 , f2 , . . . , fk } and take φ(f ) = 1,

φ(fj ) = 0

for j = 1,

φ(T) = 0.

Now, H 2 (G; Z) ∼ = Hom(H2 (G), Z), so φ is an element of degree 2 cohomology. Moreover, the induced homomorphism ∂ ∗ : H 2 (G; Z) → H 2 (S3 ; Z) has the following effect on the generator x ∈ H2 (S3 ; Z) ∼ = π2 (S3 ) = Z: ∂ ∗ (φ)(x) = φ(∂∗ (x)) = φ(f ) = 1, ¯ where x¯ ∈ H 2 (S3 ; Z) = Hom(H2 (S3 ), Z) which implies that ∂ ∗ (φ) = x, is defined by x(x) ¯ = 1. But, from Example B.11, we see that the cup product powers x¯ m are nonzero for all m ≥ 1. Thus 0 = x¯ m = (∂ ∗ (φ))m = ∂ ∗ (φ m ) which implies that φ m = 0 for all m in the finite dimensional manifold G. This contradiction then shows that f = 0. Hence H2 (G) = T, the torsion part, and H 2 (G; R) = 0. Remark 1.82 In fact, a theorem due to S. Weingram allows us to show the stronger result π2 (G) = 0. Weingram’s theorem (see [263]) says that a map f : S2n+1 → K(A, 2n) which has 0 = f∗ : H2n (S2n+1 ; Z) → H2n (K(A, 2n); Z) is incompressible; that is, f does not factor through a finite complex. In Proof 3 of Theorem 1.45, if the classifying map τ : S3 → BG represents a torsion element of π2 (G) = H2 (G), then

41

42


for some pr , where p is a prime, there is a dual cohomology class τ ∈ Hom(H2 (G), Z/pr ) ⊆ H 2 (G; Z/pr ). Because, in general, H k (X; A) ∼ = [X, K(A, k)] (where [X, K(A, k)] denotes the set of homotopy classes of maps), we obtain the following homotopy commutative diagram by focusing on the connecting map in the Puppe sequence S3 B BB BB BB B! ∂

φ

G

/ K(Z/pr , 2) : uu uu u u uu τ uu

Here, with ι ∈ H 2 (K(Z/pr , 2); Z/pr ) the fundamental class, we have φ ∗ (ι) = τ ∗ (ι) = ∂ ∗ τ , so ∂ ∗ pulls back τ ∈ H 2 (G; Z/pr ) into H 2 (S3 ; Z/pr ). By ∂ ∗ the same argument as before, this class in H 2 (S3 ; Z/pr ) is nontrivial. By Weingram’s theorem, this is impossible because G is a compact manifold. Therefore, no such class τ can exist and π2 (G) = 0.

1.12

Stiefel and Grassmann manifolds

We will now define Stiefel and Grassmann manifolds and compute their cohomology. We deduce from this determination the universal fibrations associated to O(n), U(n), Sp(n). In Proposition 1.87, a particular case of a “model of a fibration” is also given. This is a prototype of the relative models developed in Chapter 2. Example 1.83 (Stiefel manifolds) We denote the Stiefel manifolds by, for 1 ≤ k < n, SO(n) O(n) = , O(n − k) SO(n − k) SU(n) U(n) = , Vn,k (C) = U(n − k) SU(n − k) Sp(n) Vn,k (H) = , Sp(n − k) Vn,k (R) =

and Vn,n (R) = O(n), Vn,n (C) = U(n), Vn,n (H) = Sp(n). As particular cases, we get Vn,1 (R) = Sn−1 , Vn,1 (C) = S2n−1 , Vn,1 (H) = S4n−1 . Stiefel manifolds Vn,k (−) are isomorphic to the spaces of orthonormal k-frames in the respective Rn , Cn , Hn .

1.12


Example 1.84 (Grassmann manifolds) We denote the Grassmann manifolds by, for 1 ≤ k < n: O(n) , O(k) × O(n − k) U(n) Gn,k (C) = , U(k) × U(n − k) Sp(n) . Gn,k (H) = Sp(k) × Sp(n − k) Gn,k (R) =

As particular cases, we get Gn,1 (R) = RP(n − 1), Gn,1 (C) = CP(2n − 1), Gn,1 (H) = HP(4n − 1). Grassmann manifolds Gn,k (−) are isomorphic to the spaces of k-dimensional subspaces in the respective Rn , Cn , Hn . Before the statement, we recall that ∧(V) (Q[V]) denotes the exterior (polynomial) algebra generated by the Q-vector space V. Such a vector space is sometimes represented by a basis, V = (x1 , . . . , xk ). Proposition 1.85 The cohomology algebras of complex and quaternionic Stiefel manifolds are described by:

H ∗ (Vn,k (C); Q) = ∧ x2(n−k)+1 , . . . , x2n−1 with x2i−1 ∈ H 2i−1 (Vn,k (C); Q);

H ∗ (Vn,k (H); Q) = ∧ y4(n−k)+3 , . . . , y4n−1 with y4i−1 ∈ H 4i−1 (Vn,k (H); Q). By using the particular cases Vn,n (C) = U(n), Vn,n (H) = Sp(n) and Theorem 1.81, one gets the following result immediately. Corollary 1.86 The unitary and symplectic groups and their classifying spaces have for cohomology algebras: H ∗ (U(n); Q) = ∧ (x1 , x3 , . . . , x2n−1 ) with x2i−1 ∈ H 2i−1 (U(n); Q); H ∗ (SU(n); Q) = ∧ (x3 , . . . , x2n−1 ) with x2i−1 ∈ H 2i−1 (SU(n); Q); H ∗ (Sp(n); Q) = ∧ (y3 , y7 , . . . , y4n−1 ) with y4i−1 ∈ H 4i−1 (Sp(n); Q); H ∗ (BU(n); Q) = Q [c1 , c2 , . . . , cn ] with ci ∈ H 2i (BU(n); Q); H ∗ (BSU(n); Q) = Q [c2 , . . . , cn ] with ci ∈ H 2i (BSU(n); Q); H ∗ (BSp(n); Q) = Q [q1 , q2 , . . . , qn ] with qi ∈ H 4i (BSp(n); Q). A locally trivial fiber bundle, with a numerable trivializing open cover of the base (see [74]), is a fibration, so the bundle satisfies the homotopy lifting property (see Section 2.5.2 and Exercise 1.9). For the total space of a fibration, Sullivan theory supplies a nice algebraic model which contains models

43

44


of the base and of the fiber. As an aperitif, we now present a particular case, sufficient for our study of Stiefel and Grassmann manifolds. Recall first, from Corollary B.13, that a bundle Sk an exact sequence, called the Gysin sequence, H j (B; Q)

δ

p∗

/ H j+k+1 (B; Q)

/ H j+k+1 (E; Q)

/E

p

/ B induces

/ H j+1 (B; Q) . . .

where δ(x) = x ∪ e for some e ∈ H k+1 (B; Q). p

/E / B be a fibration such that the class Proposition 1.87 Let S2n+1 2n+2 e∈H (B; Q) appearing in the Gysin sequence is zero. Then we have an isomorphism of algebras

H ∗ (B; Q) ⊗ ∧y

∼ =

/ H ∗ (E; Q),

where y is of degree 2n + 1. Proof The Gysin exact sequence splits into short exact sequences: 0

/ H j (B; Q)

p∗

/ H j (E; Q)

σ∗

/ H j−2n−1 (B; Q)

/ 0.

We choose a class y ∈ H 2n+1 (E; Q) such that σ ∗ (y) = 1 and define a cochain map : H ∗ (B) ⊗ ∧y → H ∗ (E) by (ω) = p∗ (ω) if ω ∈ H ∗ (B) and (y) = y. Because y is of odd degree, the map is an isomorphism of algebras. Remark 1.88 In [112, page 320], the class e is defined as the Euler class of the bundle. In the case of a sphere bundle, this class coincides with the pullback of the Thom class along the zero section (see [41, Theorem 13.2, page 390]). A third equivalent definition of Euler class will be given in Example 1.96. Proof of Proposition 1.85 Let n ≥ 1 be an integer. For k = 1, we have Vn,1 (C) = S2n−1 and the statement is true. We now use induction on k by supposing that the formula is true for k − 1. We consider the fibration S2(n−k)+1

/ Vn,k (C)

/ Vn,k−1 (C) .

By the induction hypothesis, the Euler class in H 2(n−k)+2 (Vn,k−1 (C); Q) is zero. The result follows directly from Proposition 1.87. The argument is similar for the symplectic Stiefel manifolds.

1.12


In the real case, the proof works also by induction on k, using the fibration Sn → SO(n + k)/SO(n) → SO(n + k)/SO(n + 1), but one needs to determine the Euler class of this fibration. We do not give the proof here, referring the reader to [199, page 121]. Proposition 1.89 The real Stiefel manifolds have for cohomology algebras:

H ∗ (Vn+k,k (R); Q) = ∧ z2n+1 , z2n+5 , . . . , z2n+2k−3 , n odd, k even,

= ∧ z2n+1 , z2n+5 , . . . , z2n+2k−5 , zn+k−1 , n odd, k odd,

= ∧ zn , z2n+3 , z2n+7 , . . . , z2n+2k−3 , n even, k odd,

= ∧ zn , z2n+3 , z2n+7 , . . . , z2n+2k−5 , zn+k−1 , n odd, k even, with zi ∈ H i (Vn+k,k (R); Q). Noting the particular cases Vn,n (R) = O(n) and using Theorem 1.81, we deduce the following. Corollary 1.90 The orthogonal groups and their classifying spaces have for cohomology algebras: H ∗ (SO(2m); Q) = ∧ (z3 , z7 , . . . , z4m−5 , z2m−1 ) , H ∗ (SO(2m + 1); Q) = ∧ (z3 , z7 , . . . , z4m−1 ) , H ∗ (BSO(2m); Q) = Q [p1 , p2 , . . . , pm−1 , χ ] , H ∗ (BSO(2m + 1); Q) = Q [p1 , p2 , . . . , pm ] , with zi ∈ H i (SO(−); Q), pi ∈ H 4i (BSO(−); Q), χ ∈ H 2m (BSO(2m); Q). Remark 1.91 The results of Proposition 1.85 and Corollary 1.86 are true for cohomology with coefficients in Z, since these cohomology groups have no torsion (see [199, page 119]). This is not the case for the real Stiefel manifolds; their cohomology has 2-torsion. Being concerned here only with coefficients in a field of characteristic zero, we refer the reader to [199, page 121] for an explicit description of the cohomology with coefficients in Z2 . As a direct consequence of the previous computations and Theorem 1.73, we obtain the following. Proposition 1.92 1. The principal O(k)-bundle, Vn+k,k (R) → Gn+k,k (R), is n-universal. 2. The principal U(k)-bundle, Vn+k,k (C) → Gn+k,k (C), is (2n + 1)universal.

45

46


3. The principal Sp(k)-bundle, Vn+k,k (H) → Gn+k,k (H), is (4n + 3)universal. Corollary 1.93 Set V∞,k = ∪n Vn+k,k and G∞,k = ∪n Gn+k,k for the fields R, C and H. Then the three following principal fibrations are universal: V∞,k (R) → G∞,k (R), for G = O(k); V∞,k (C) → G∞,k (C), for G = U(k); V∞,k (H) → G∞,k (H), for G = Sp(k). Remark 1.94 For the universal fibrations associated to SO(n) and SU(n), it is necessary to introduce the spaces of oriented vector subspaces: SGn,k (R) = O(n)/SO(k) × O(n − k) and SGn,k (C) = U(n)/SU(k) × U(n − k). See [145, Theorem 6.1 of Chapter 4]. Definition 1.95 Let p : E → B be a principal G-bundle with classifying map ψ : B → BG. A characteristic class of p is an element of ψ ∗ H ∗ (BG; R) ⊆ H ∗ (B; R), for a commutative ring R. Since we work with coefficients in Q, we consider only rational characteristic classes. Example 1.96 We list here the characteristic classes corresponding to the different Lie groups of matrices. • For a U(n)-bundle, the characteristic classes are generated by the image

of the classes ci ∈ H ∗ (BU(n); Q) = Q[c1 , . . . , cn ]. These are called Chern classes. • For a SO(2m + 1)-bundle, the characteristic classes are generated by the image of the classes pi ∈ H ∗ (BSO(2m + 1); Q) = Q[p1 , . . . , pm ]. These are called Pontryagin classes. • For a SO(2m)-bundle, the characteristic classes are generated by the image of the Pontryagin classes just defined, and the universal Euler class χ ∈ H 2m (BSO(2m); Q). • For Sp(n), the characteristic classes are generated by the image of the classes qi ∈ H 4i (BSp(n); Q). These are called symplectic Pontryagin classes. When the bundle Sr → E → B has SO(r + 1) as structural group, the notion of Euler class introduced in Remark 1.88 from the Gysin sequence may be identified with the definition of this section (see [145, Proof of Theorem 6.5, page 103]). In the case of S1 = SO(2), one has BS1 = CP∞ = K(Z, 2) and the Euler class with coefficients in Z is the image of the generator

1.13

The Cartan–Weil model

of H 2 (K(Z, 2); Z) (see Section 6.5 for a description of the relation between the Euler class and the flatness of the associated vector bundle). One can also observe some relations between characteristic classes coming from different inclusions between Lie groups (see [198, page 176] for a description of them). For instance, the canonical inclusion U(m) ⊂ SO(2m + 1) gives a map → BSO(2m + 1) which induces in ψ : BU(m) ∗ r cohomology ψ (pj ) = r+s=j (−1) cr cs ∈ H 4j (BU(m) [199, page 144].

1.13


Once the rational cohomology of Lie groups was determined, calculating the cohomology of general homogeneous spaces became an important goal. The combined work of Cartan, Chevalley and Weil after World War II provided an algebraic model whose cohomology was the cohomology of the homogeneous space under consideration. This is the genesis of the theory of minimal models which is the subject of Chapter 2. Indeed, we shall give the minimal model version of the Cartan–Weil model in Theorem 2.71. Here we want to both generalize and particularize the Cartan–Weil model for homogeneous spaces. We generalize the model by describing an algebraic model for an action of a Lie group on a space. We particularize by taking the simplest example: the action of a circle on a manifold. The advantage of the original Cartan–Weil model over newer minimal models is that we can often see more geometry in the Cartan–Weil model because it is constructed from forms. (In particular, there are models based on harmonic forms and the Hodge decomposition which often are more computable; see, for instance, [11].) Our treatment is essentially that of [15]. In Theorem 1.28, we saw that, if a compact connected Lie group G acts on a closed manifold M, then H ∗ (M; R) = HL∗ (M), where HL∗ (M) is the cohomology of the complex of left invariant forms. While this is an interesting result, there is a more important cohomology associated to a group action. This is the equivariant cohomology defined as follows. As we saw above, there is a universal principal bundle G → EG → BG with a free right action of G on EG and BG = EG/G. If G also acts on M, then there is a free right action of G on EG × M given by g(e, m) = (eg −1 , gm). The quotient is then def

MG = EG ×G M = (EG × M)/G, and there is a fibration (called the Borel fibration), p

M → MG → BG where p([e, m]) = [e].

47

48


Definition 1.97 The equivariant cohomology of M with respect to an action of G on M is defined to be ∗ HG (M) = H ∗ (MG ).

Any coefficients may be used in the definition. For instance, we will be ∗ (M; Q) or H ∗ (M; R). The equivariant cohomology interested in either HG G is important because it gives information not only about M, but about the G-action also. While the Borel fibration will be studied extensively from topological and algebraic perspectives in Section 7.2, here we want to hint at how Lie group actions lead to topology and, especially, algebraic models. Since we are only interested here in motivating the models of Chapter 2, let’s take a special case of a smooth action of S1 on a closed manifold M. For a fixed m ∈ M, the orbit map S1 → M given by g → gm induces a linear map Te S1 → Tm M with Xθ → Xm , where Xθ is a basis for Te S1 = R, the Lie algebra of S1 . Doing this for every m ∈ M produces a vector field X on M called the fundamental vector field of the action. As such, we can form the operators i(X) and L(X) (see Section A.2). The Weil algebra W of the circle is W = ∧(θ, u) = Exterior(θ) ⊗ Polynomial(u), where θ is in degree 1 and u is in degree 2. A differential D is defined on W by declaring Dθ = u and Du = 0. Notice that the cohomology of this differential graded algebra is zero (except for H 0 (W) = R). We mention here that the Weil algebra can be defined for other Lie groups also using the structure constants of the associated Lie algebra to describe the differential. We can extend the action of the operators i(X) and L(X) from the de Rham algebra ADR (M) to the complex W ⊗ ADR (M) as follows. On W, define i(X)θ = 1,

i(X)u = 0,

L(X)θ = 0,

L(X)u = 0.

In fact, the general definition for L(X) is L(X) = i(X)D + Di(X) and the equations above hold in the S1 case. Definition 1.98 The basic subcomplex B (M) ⊂ W ⊗ ADR (M) consists of all elements α with i(X)α = 0 and L(X)α = 0. Of course, a similar definition holds for general compact connected G. The fundamental result that we shall not prove in the present framework is (in its general form, see [15]),

1.13


Theorem 1.99 If there is a smooth action of a connected compact Lie group G on a closed manifold M, then ∗ (M; R). H ∗ (B (M)) ∼ = HG

So B (M) is, in some sense, a model for the total space of the Borel fibration, MG . Clearly, however, it is not a particularly tractable model. Below, we shall find a better and more intuitive model. 1 for simplicity. Any element of W⊗ADR (M) Let’s specialize to an S -action ka + j has the form ω = u k k j θu bj where k and j are non-negative integers and ak , bj ∈ ADR (M). The next result identifies the basic elements. Proposition 1.100 Using the notation above, an element ω ∈ W ⊗ ADR (M) is basic if and only if L(X)ak = 0 and i(X)ak = −bk for each k. Proof First, recall that we extended i(X) and L(X) to W ⊗ ADR (M) above. Now, since i(X) and L(X) are graded derivations, we have i(X)ω = uk (i(X)ak ) + (uj bj − θuj i(X)bj ) = 0, j

k

L(X)ω =

k

uk (L(X)ak ) +

θ uj (L(X)bj ) = 0.

j

Now note that the uk and uk θ are algebraically independent over ADR (M). Hence (using the standard facts that i(X)2 = 0 and i(X)L(X) = L(X)i(X)), the equations above are seen to be equivalent to those of the statement of the proposition. Now we can begin to simplify our model for equivariant cohomology. While the basic forms tell us what we need to know, they are unwieldy because of the condition i(X)α = 0. We might reasonably ask what this has to do with the action. The forms that should be important are the forms that are invariant under the flow along orbits. These are, of course, identified with the forms that are annihilated by L(X). Therefore, let X = {α ∈ ADR (M) | L(X)α = 0}. As we have said, these are the forms that are invariant under the circle action. Adjoin a degree 2 generator u to X to obtain an algebra X [u] = X ⊗ R[u] and an algebra homomorphism φ : X [u] → W ⊗ ADR (M) defined by φ(α) = α − θi(X)α and φ(u) = u. Lemma 1.101 The image of φ lies in the basic subalgebra B (M). Proof We must show that L(X)φ(α) = 0 and i(X)φ(α) = 0, for α ∈ X . We compute (again using the extensions of i(X) and L(X) to W ⊗ ADR (M)

49

50


and the relations i(X)2 = 0, L(X)i(X) = i(X)L(X)).

L(X)φ(α) = L(X)(α − θ i(X)α) = L(X)α − (L(X)θ)i(X)α − θ L(X)i(X)α = 0 − 0 − θi(X)L(X)α = 0, where we have used the fact that L(X)α = 0 since X consists of invariant forms. Similarly, since i(X)θ = 1, we have i(X)φ(α) = i(X)(α − θ i(X)α) = i(X)α − (i(X)θ)i(X)α + θi(X)i(X)α = i(X)α − i(X)α + 0 = 0.

It is not difficult to see that, in fact, the equations of Proposition 1.100 imply that φ is an algebra isomorphism onto its image, ∼ =

φ : X [u] → B (M). Therefore, we obtain a differential dX on X [u] by transporting the differential on W ⊗ ADR (M) back via φ −1 . By definition, we have φdX = Dφ, where D is a differential defined as follows. On the Weil algebra, D is the differential we defined above: Dθ = u and Du = 0. On ADR (M), D is simply the exterior derivative d. Now let’s compute to see what dX has to be. Suppose α ∈ X . Using L(X)α = di(X)α + i(X)dα = 0, we obtain the following. Dφ(α) = D(α − θ(i(X)α)) = dα − u(i(X)α) + θd(i(X)α) = dα − ui(X)α − θi(X)dα = φ(dα − ui(X)α), since φ(dα) = dα − θi(X)dα and φ(ui(X)α) = φ(u)φ(i(X)α) = u(i(X)α − θi(X)i(X)α) = ui(X)α. Therefore, in order for Dφ = φdX to hold, we should define dX (u) = 0 and, for α ∈ X , dX α = dα − ui(X)α. Thus we see that φ : (X [u], dX ) → (B (M), D) is an isomorphism of differential graded algebras. Hence, we have the following result.

1.13


Theorem 1.102 Suppose G = S1 acts smoothly on a closed manifold M. The Cartan–Weil model is the complex (X [u], dX ), where X is the subcomplex of ADR (M) consisting of G-invariant forms, u is a degree 2 generator and the differential is defined by dX u = 0 and dX α = dα − ui(X)α, for α ∈ X . The Cartan–Weil model is isomorphic to the basic complex (B (M), D) of W ⊗ADR (M) and this isomorphism induces an isomorphism H ∗ (X [u], dX ) ∼ = H ∗ (B (M), D) ∼ = H ∗ (MS1 ). Thus, the Cartan–Weil model calculates the equivariant cohomology associated to the action of G on M. Remark 1.103 The equation dX α = dα − ui(X)α says something interesting. Although α is an invariant form, even if it is closed, this does not mean that α is equivariantly closed (i.e. dX -closed). For that, we also require i(X)α = 0 – and this is exactly a basic form in the model B (M). Remark 1.104 The models can be directly related to the geometry of the Borel fibration. The homomorphism X [u] → ADR (M) obtained by evaluating u = 0 is a model for the fiber inclusion M → MS1 . Integration over the fiber is expressed on X [u] by ak uk → ( ak )uk . Remark 1.105 The whole discussion above extends to the action of a torus T n on a closed manifold M. In this case the model for MT n is ∗X1 ,...,Xn [u1 , . . . , un ], where the Xi are fundamental vector fields corresponding to a basis for the Lie algebra Rn of T nand the ui are all in degree 2. The differential dX is given by dX α = dα − k uk i(Xk )α. The differential graded algebra (X [u], dX ) is a “model” for the minimal models we shall describe in Chapter 2. Indeed, we shall see a distinct resemblance to the relative minimal models that describe the rational homotopy structure of fibrations. Namely, the algebra (X , d) comes from the manifold M (while also including information about the action), the degree 2 generator u comes from BS1 since H ∗ (BS1 ; R) = R[u] and the differential dX on X [u] is a perturbation of the differential d by a term involving the action. This is a way to construct a model of the total space of the Borel fibration M → MS1 → BS1 starting from the base and fiber; indeed, these are the types of ingredients found in the theory of minimal models that we will present in the next chapter and that will occupy and intrigue us throughout this book.

51

52


Exercises for Chapter 1 Exercise 1.1 (1) Prove the existence of the following isomorphisms of Lie groups: Spin(3) ∼ = S3 × S3 , Spin(5) ∼ = Sp(2), Spin(6) ∼ = S3 , Spin(4) ∼ = SU(4), SO(2) ∼ = Sp(1). = S1 , SU(2) ∼ Hint: [199, page 82–84]. (2) Prove the existence of homeomorphisms of topological spaces: SO(3) ∼ = S3 × SO(3). = RP(3), SO(4) ∼ Exercise 1.2 The special linear group, SL(n, R), is the subgroup of Gl(n, R) formed by the isomorphisms of determinant 1. (1) Show that SL(n, R) is a Lie group of dimension n2 − 1 and has for Lie algebra the set, sl(n, R), of n × n matrices of trace 0. (2) Prove that SO(n, R) is the maximal compact subgroup of SL(n, R). In fact, show SO(n, R) × R

n(n−1) +n−1 2

∼ = SL(n, R).

(3) Define the group SL(n, C) and prove: • SL(n, C) is a Lie group of dimension 2n2 − 2; • SU(n) × Rn

2 −1

∼ = SL(n, C).

Hint: see [113, Chapter 2] and [199, page 34]. Exercise 1.3 The real symplectic group is defined as Sp(n; R) = A ∈ Gl(2n, R) | tAJn A = Jn , 0 −In . where Jn = In 0 (1) Show that Sp(n, R) is a Lie group of dimension n(2n + 1) and has for Lie algebra the set, sp(n, R), of 2n × 2n real matrices A such that Jn AJn−1 = − tA. (2) Prove that U(n) is the maximal compact subgroup of Sp(n, R); U(n) × Rn(n+1) ∼ = Sp(n, R). (3) Define the group Sp(n, C) and prove: • Sp(n, C) is a Lie group of dimension 2n(n + 1); • Sp(n) × Rn(2n+1) ∼ = Sp(n, C); • Sp(n) ∼ = Sp(n, C) ∩ U(2n). Hint: see [113, Chapter 2] and [199, page 34]. Comment: What we call the symplectic group, denoted by Sp(n), is the orthogonal group of the quaternions. The real and complex symplectic groups, Sp(n, R) and

1.13


Sp(n, C), are the groups of isomorphisms which respect a symplectic form. Since we are mainly interested in compact Lie groups, these last two will not appear in the rest of the book and there will be no confusion in terminology. Exercise 1.4 Determine the maximal torus and the Weyl group of the spinor groups, Spin(n). Hint: see [145, 14-8]. Exercise 1.5 Let G be a compact connected Lie group. Denote by Z(g)e the connected component of the centralizer of g ∈ G containing the neutral element. (1) Show that Z(g)e is the union of the maximal tori containing g. (2) An element g ∈ G is called regular if g belongs to just one maximal torus. Otherwise, it is called singular. Show that g is regular if and only if the dimension of Z(g) is equal to the rank of G. Hint: see [199, page 267]. Comment: From the existence of a smooth map f : M → G such that dim M = dim G − 3 and such that f (M) is the set of singular elements of G (this set is not a manifold!), one can deduce that π2 (G) = 0; see [135, Theorem 4.7, page 260] or [113, Problem 34-35, page 108]. Exercise 1.6 Let G be a compact connected Lie group and let K be a closed connected subgroup of G. Let σ be a smooth automorphism of G such that σ 2 = id. Denote by (Gσ )e the connected component of the fixed point set Gσ of σ containing the unit e. The pair (G, K) is called a symmetric pair if (Gσ )e ⊂ K ⊂ Gσ . In this case, the quotient G/K is called a symmetric space. (1) Show that (U(p + q), U(p) × U(q)) is a symmetric pair. Give other examples. (See [199, page 147].) (2) Show that a symmetric space is parallelizable. (3) Let (G, K) be a symmetric pair. The group G acts on the left on G/K. Denote by g and k the Lie algebras of G and K respectively and by k⊥ the orthogonal complement of k in g∗ . Show that the cohomology H ∗ (G/K; R) is isomorphic, as an algebra, to the invariant set of ∧k⊥ . Hint: in the proof for a semisimple Lie group, replace the inverse map by the automorphism σ . Exercise 1.7 Let G be a compact connected Lie group with Lie algebra g. Recall the existence of an isomorphism χ : L (G) ∼ = ∧g∗ defined by χ(ω) = ωe . If X is a left invariant vector field, we also know that L (G) is stable by the coboundary d, the interior multiplication i(X) and the Lie derivative L(X). Let ∈ ∧p g∗ and let (X, X0 , X1 , . . . , Xp ) be elements of g. We set: (i(X))(X1 , . . . , Xp−1 ) = (X, X1 , . . . , Xp−1 ). (θ(X))(X1 , . . . , Xp ) = −

p

(X1 , . . . , [X, Xi ], . . . , Xp ).

i=1

(δ)(X0 , . . . , Xp ) =

ˆ i, . . . , X ˆ j , . . . , Xp ). (−1)i+j ([Xi , Xj ], X0 , . . . , X

i n; • πq (fn ) is an isomorphism for q ≤ n; • pn : Xn → Xn−1 is a principal fibration obtained as a pullback of the

path fibration K(πn (X), n) = (K(πn (X), n+1)) → PK(πn (X), n+1) → K(πn (X), n + 1) along a map kn : Xn−1 → K(πn (X), n + 1).

The sequence of fibrations is called the Postnikov tower of X and the maps kn are called the associated k-invariants ([240]). Let (∧V, d) be the minimal model of X. We then have the following properties. • The minimal model of Xn is given by the sub-cdga (∧V ≤n , d). • The minimal models of fn and pn are, respectively, given by the injections

(∧V ≤n , d) → (∧V, d) ,

(∧V ≤n−1 , d) → (∧V ≤n , d) .

• The minimal model of the nth k-invariant kn : Xn−1 → K(πn (X),

n + 1) is given by the map kn : (∧(s−1 V n ), 0) → (∧V ≤n−1 , d), where (s−1 V n )n+1 = V n , (s−1 V n )q = 0, q = n + 1, kn (s−1 v) = dv.

The Lusternik–Schnirelmann category of a space X, cat X, is the least integer n such that X can be covered by n + 1 open sets, each contractible in X. For instance the category of a point is zero, and the category of a sphere is one. The properties of the Lusternik–Schnirelmann category, its description in terms of minimal models and its role in algebraic and differential topology are described in [66] and [87]. Here we only recall the main points. A lower bound for cat X is given by the rational cup length cup0 (X) = max { n | ∃ α1 , . . . , αn ∈ H + (X; Q) such that α1 · · · αn = 0 } , and an upper bound is given by the dimension of X. cup0 (X) ≤ cat X ≤ dim X .

2.5

Minimal models and homotopy theory

Suppose now that X is simply connected. By a result of Toomer ([253]), the category of the rationalization X0 (see Subsection 2.6.1) is less than or equal to the category of X, cat X0 ≤ cat X . The integer cat X0 is called the rational category of X, and is denoted cat0 (X). Because the rational homotopy type of X is encoded in its minimal model, it is no surprise that this invariant can be calculated from the minimal model (∧V, d) of X. Theorem 2.80 ([86]) The integer cat0 (X) is the least integer m such that the ¯ admits minimal model of the projection qm : (∧V, d) → (∧V/(∧>m V), d) a homotopy retraction. Denote by ρm : (∧V, d) → (∧Wm , d) the minimal model of qm . ρm

/ (∧Wm , d) (∧V, d) S SSS SSS SSS SSS SS) qm ¯ (∧V/ ∧>m V, d)

The rational category of X is the least integer m such that there is a morphism r : (∧Wm , d) → (∧V, d) with r ◦ ρm = id∧V . One of the main results concerning rational category is the so-called mapping theorem. Theorem 2.81 (The mapping theorem [87, Theorem 28.6]) Let f : X → Y be a continuous map between nilpotent spaces. If π∗ (f ) ⊗ Q is injective, then cat0 (X) ≤ cat0 (Y). The rational homotopy Lie algebra of a space of finite rational category satisfies properties that are similar to the properties of finite dimensional spaces. Theorem 2.82 ([87, Theorem 33.6 and 36.8]) Let X be a simply connected space with finite Betti numbers and finite rational category. When dim π∗ (X) ⊗ Q < ∞, then dim H ∗ (X; Q) < ∞ and X is a rationally elliptic space. When dim π∗ (X) ⊗ Q = ∞, then 1. There is B > 1 such that ki=1 rank πi (X) > Bk for k enough large. 2. dim πodd (X) ⊗ Q = ∞. 3. There is an integer N such that for all α ∈ πq (X) ⊗ Q with q > N, there is β ∈ π∗ (X) ⊗ Q such that ad(α)n (β) = 0 for all n.

89

90

2 : Minimal models

Minimal models are also a very good tool for describing the rational homotopy type of the homotopy cofiber of a map. If f : X → Y is a continuous map, its homotopy cofiber Cf is the space Y ∪ CX/ ∼, where CX = X × [0, 1]/X × {1} and where ∼ identifies (x, 0) ∈ CX to f (x) ∈ Y. Recall that if f denotes the inclusion of a sub-CW complex or the inclusion of a submanifold, then Cf is homotopy equivalent to the usual quotient Y/X. Denote by Mf the minimal model of f and suppose the following diagram is commutative up to homotopy, where the vertical arrows are quasi-isomorphisms.

MY

(A, d)

Mf

g

/ MX

/ (B, d).

Then we have, Theorem 2.83 ([87, Proposition 13.6]) If g is surjective, then the minimal model of Q ⊕ Ker g is the minimal model of Cf . In particular the geometric realization of the minimal model of Q ⊕ Ker g is the rationalization of Cf .

2.6

Realizing minimal cdga’s as spaces

In order to apply the algebra of minimal cdga’s to geometry, we have to understand how algebraic data can be realized geometrically. We begin by considering how a minimal cdga may be realized topologically.

2.6.1

Topological realization of a minimal cdga

Consider on one side the category of nilpotent spaces with finite Betti numbers TopN and, on the other side, the category AN composed of the cdga’s (A, d) with H 0 (A, d) = Q, and which admit a finite type minimal model. There is a realization functor going from AN to TopN (see [38] and [246]) that has the following properties: • The realization of a minimal cdga (∧V, d), (∧V, d), is a rational space;

that is, its homotopy groups are rational vector spaces. MX and (∧V, d) → (∧V, d) are inverse to each other up to rational homotopy equivalence. The minimal model of

• The correspondences X →

2.6

Realizing minimal cdga’s as spaces

(∧V, d) is (∧V, d); moreover, there is a map X → MX that induces an isomorphism in rational cohomology. The space MX is called the rationalization of X, and is denoted by X0 . This space is characterized by the following properties: 1. X0 is a rational space. 2. There is a map f : X → X0 inducing an isomorphism in rational cohomology. 3. If Y is a rational space and g : X → Y is a continuous map, then there is a map, unique up to homotopy, h : X0 → Y such that g h ◦ f . The correspondence between AN and TopN extends to maps. • If X, Y ∈ TopN , then rationalization induces a bijection

[X, Y] → [X0 , Y0 ] ∼ = [MY , MX ] . • The realization of a relative minimal cdga (∧V, d) → (∧V ⊗ ∧W, D) is

a fibration p : E → B.

The rationalization of a nilpotent finite type CW complex is not a finite type CW complex because the homotopy groups of the rationalization are Q-vector spaces, while the homotopy groups of a finite type CW complex are finitely generated. However, the realization described above can, in general, be improved, in a nonfunctorial way, to give a finite type CW complex whose minimal model is the original one. • Each finite type minimal cdga (∧V, d) is the minimal model of a nilpotent

space X with finite Betti numbers. If H ∗ (∧V, d) is finite dimensional, then we can choose X to be a finite CW complex. • Each map between finite type minimal cdga’s is the minimal model of a continuous map between finite type CW complexes. • Each finite type relative minimal cdga (∧V ⊗ ∧W, D) is the relative minimal model of a fibration p : E → B where E and B are finite type CW complexes.

2.6.2

The cochains on a graded Lie algebra

Let L be a graded Lie algebra. We can associate to L its cochain algebra C ∗ (L) = (∧V, d) with V p+1 ∼ = Hom(Lp , Q) and differential d = d1 defined by x, s[f , g] = (−1)|g| d1 x, sf ∧ sg , The next proposition follows directly.

x ∈ V, f , g ∈ L .

91

92

2 : Minimal models

Proposition 2.84 Every finite type graded Lie algebra L such that L0 = L1 = 0 can be realized as the rational homotopy Lie algebra of a simply connected space. The cochain algebra (∧V, d) admits a bigradation defined by (∧V)p,q = (∧p V)q−p . Since d(∧V)p,q ⊂ (∧V)p+1,q , this bigradation induces a bigradation on the cohomology. The qth rational cohomology vector space of a graded Lie algebra L, H q (L; Q), is by definition the graded vector space H q,∗ (C ∗ (L)). Every short exact sequence of Lie algebras 0 → L1 → L2 → L3 → 0 induces a relative minimal cdga C ∗ (L3 ) → C ∗ (L2 ) → C ∗ (L1 ) , and is therefore realized by a fibration. A minimal presentation of a graded Lie algebra is a short exact sequence f

0 → L(W) → L(S) → L → 0 where L( ) denotes the free graded Lie algebra functor and where f (W) ⊂ L≥2 (S). By [171] and [248], we then have isomorphisms of graded vector spaces S∼ = H 1 (C ∗ (L)) ,

W∼ = H 2 (C ∗ (L)) .

When L is a finite dimensional nilpotent Lie algebra concentrated in degree 0, the cochain algebra C ∗ (L) is finite dimensional. Its realization is a nilmanifold. The theory of nilmanifolds will be described in Section 3.2.

2.7

Formality

Definition 2.85 A nilpotent space X, with minimal model (∧V, d), is called formal if there is a quasi-isomorphism ϕ : (∧V, d) → (H ∗ (X; Q), 0) . Remark 2.86 • We can also define a cdga (A, d) to be formal if there is a chain of quasi-

isomorphisms (A, d) ← (B1 , d1 ) → · · · (Bk , dk ) → (H ∗ (A), 0) .

2.7

Formality

We can take the minimal models of (A, d), the minimal models of the (Bi , di ) and the minimal models of the morphisms by Theorem 2.26. By applying Corollary 2.13, we then see that the condition above is equivalent to Definition 2.85. • By a result of Sullivan (see Proposition 2.101, [206] and [246]), the definition of formality given above is equivalent to the existence of a quasi-isomorphism over the reals, ϕ : (∧V, d) ⊗Q R → (H ∗ (X; R), 0) . For a manifold M, this is also equivalent to the existence of a sequence of quasi-isomorphisms of R-cdga’s connecting ADR (M) and (H ∗ (M; R), 0). We will come back to this point in Proposition 2.101. For a classical criterion guaranteeing formality, see Exercise 2.3. It follows directly from the definition that Lie groups, complex and quaternionic Stiefel manifolds and loop spaces are formal spaces because their respective cohomologies are isomorphic to their minimal models. Example 2.87 (Product of formal spaces) Let X and Y be formal spaces with respective minimal models (∧V, d) and (∧W, d). By the formality of X and Y, we have quasi-isomorphisms ϕ : (∧V, d) → (H ∗ (X; Q), 0)

and

ψ : (∧W, d) → (H ∗ (Y; Q), 0) .

The minimal model of X × Y is the tensor product (∧V, d) ⊗ (∧W, d) and we have a quasi-isomorphism ϕ ⊗ ψ : (∧V, d) ⊗ (∧W, d) → (H ∗ (X; Q) ⊗ H ∗ (Y; Q), 0) ∼ = (H ∗ (X × Y; Q), 0) . This shows that X × Y is formal. Example 2.88 (Retract of a formal space) If X is a formal space and Y is a retract of X, then Y is a formal space. Denote i the injection and r the retraction i

r

Y → X → Y,

r ◦ i idY .

By the formality of X, we have a quasi-isomorphism ϕ : MX → (H ∗ (X; Q), 0). Then the following composition of morphisms is a quasiisomorphism

MY

Mr

/ MX

ϕ

/ (H ∗ (X; Q), 0)

H ∗ (i;Q)

/ (H ∗ (Y; Q), 0) .

93

94

2 : Minimal models

This shows that Y is a formal space. In particular, if a product X × Y is formal, the spaces X and Y are formal. An important property of formal spaces is the vanishing of all their Massey products. Let’s recall the definition. Definition 2.89 Let (A, d) be a cdga with cohomology H ∗ . Let a, b, c be cohomology classes whose products ab and bc are zero. Choose cocycles (or closed forms if we work with the de Rham complex) x, y and z representing a, b and c. By definition, there are elements v and w such that dv = xy and dw = yz. The element vz − (−1)|x| xw is then a cocycle whose cohomology class depends on the choice of v and w. The set a, b, c formed by the cohomology classes constructed using all the possible choices of v and w is called the triple Massey product of a, b and c. The triple Massey product is said to be trivial if the element 0 belongs to the set a, b, c. Denote by I the ideal of H ∗ generated by the classes of a and c. The set a, b, c projects to a single element in H ∗ /I. Moreover, this element is 0 if and only if the triple Massey product is trivial. Proposition 2.90 Let X be a formal space. Then all triple Massey products vanish. Proof Denote by (∧V, d) the minimal model of X. By definition we have a quasi-isomorphism ϕ : (∧V, d) → (H ∗ (X; Q), 0) . Now take three cohomology classes a, b and c represented by the cocycles x, y and z in (∧V, d), and suppose that there are elements v and w such that dv = xy and dw = yz. The cocycle vz − (−1)|x| xw is mapped by ϕ to a cocycle in the ideal generated by the elements ϕ(x) = a and ϕ(z) = c. Since ϕ is a quasi-isomorphism, this implies that the triple Massey product set a, b, c belongs to the ideal generated by a and c. This Massey product is thus zero. Example 2.91 (A nonformal manifold) Denote by q : S2 ×S2 → S4 the map obtained by collapsing the wedge S2 ∨ S2 to a point, and take the pullback of the Hopf fibration S3 → S7 → S4 along q (see Example 2.68). We obtain in this way a principal bundle S3 → M → S2 × S2 , whose relative minimal model is given by (∧(a, b, u, v), d) → (∧(a, b, u, v, t), d)

2.7

Formality

with |a| = 2, |b| = 2, |u| = 3, |v| = 3, |t| = 3, da = 0, du = a2 , db = 0, dv = b2 , dt = ab. We can compute the cohomology in low degrees. A basis of H 2 (M; Q) is given by the classes a and b; H 3 (M; Q) = H 4 (M; Q) = 0. The cohomology in degree 5 has dimension two, and a basis is given by the two nontrivial Massey products a, a, b = [ub − at] ,

and a, b, b = [tb − av] .

Because these Massey products do not vanish, the manifold M is not formal. Definition 2.92 (Higher order Massey products) Let u1 , . . . , up be cohomology classes. A defining system for the Massey product u1 , . . . , up is a collection of cochains mij , 1 ≤ i ≤ j ≤ p, (i, j) = (1, p), with mii a cocycle representative for ui , i = 1, . . . , p and d(mij ) =

j−1 (−1)|mik | mik · mk+1,j . k=i

We then form the cocycle p−1

α{mij } =

(−1)|m1k | m1k · mk+1,p .

k=1

The Massey product is trivial if there is a system mij for which the cohomology class of α{mij } is zero.

2.7.1

Bigraded model

The main tools for the study of formality of spaces and manifolds are bigraded and filtered models (see [132]). Theorem 2.93 (Bigraded model of a graded commutative algebra) Let A be a finitely generated graded commutative algebra. We suppose that A0 = lk. Then the cdga (A, 0) admits a minimal model ϕ : (∧V, d) → (A, 0) where V is equipped with a lower gradation V = ⊕p≥0 Vp extended in a multiplicative way to ∧V and where the following properties hold. 1. d(Vp ) ⊂ (∧V)p−1 . In particular, d(V0 ) = 0. Therefore the cohomology is a bigraded algebra H ∗ (∧V, d) = ⊕p≥0 Hp (∧V, d). 2. Hq (∧V, d) = 0 for q > 0, and H0 (∧V, d) ∼ = A. 3. ϕ(Vp ) = 0 for p > 0. The cdga (∧V, d) is called the bigraded model of the graded algebra A. The construction of the bigraded model begins with the generators (or indecomposables) of A, V0 = A+ /(A+ · A+ ). We define d|V0 = 0 and

95

96

2 : Minimal models

ϕ : ∧ V0 → A by extending a vector space splitting V0 → A. Thus, at the very first stage, we obtain a surjection from ∧V0 to A. We form ∧(V0 ⊕ V1 ) and define the differential on V1 to kill the kernel of ∧V0 → A. Note that d(V1 ) ⊂ ∧V0 . The addition of V1 may introduce yet another kernel in cohomology and this must be killed by adding in V2 with d(V2 ) ⊂ (∧(V0 ⊕ V1 ))1 . The process goes on, eventually resulting in the bigraded model. When X is a formal space, the minimal model of X is the minimal model of the cdga (H ∗ (X; Q), 0), so we can therefore choose the bigraded model of the algebra H ∗ (X; Q) as a minimal model for X. Proposition 2.94 Let M be a formal nilpotent manifold. Then all the Massey products u1 , . . . , up , p ≥ 3, are trivial in the de Rham complex ADR (M). Proof Consider the Massey product u1 , · · · , up with p ≥ 3. In the bigraded model (∧V, d) we can choose a defining system mij with mij ∈ (∧V)j−i . This shows that the element α{mij } is a cocycle in (∧V)p−2 and therefore is a coboundary because of Theorem 2.93 (3). Now, since (∧V, d) is the minimal model of M, we have a quasi-isomorphism ψ : (∧V, d) → ADR (M) and the family ψ(mij ) is a trivial defining system for the Massey product in ADR (M). This shows that all Massey products are zero.

2.7.2

Obstructions to formality

Let (A, dA ) be a cdga. We suppose that H 0 (A, dA ) = Q and that each H p (A, dA ) is finite dimensional. We denote by µ : (∧V, d) → (H ∗ (A, dA ), 0) a bigraded model of the cohomology. There then exists a perturbation of (∧V, d) that is a (not necessarily minimal) Sullivan model of (A, dA ). More precisely, we have the following. Theorem 2.95 ([132]) With the previous notation, there is a differential D on ∧V such that 1. (D − d)(Vp ) ⊂ (∧V)≤p−2 (In other words the differential can be written D = d + d2 + d3 + · · · where dq (Vp ) ⊂ (∧V)p−q ). 2. (∧V, D) is a Sullivan model for (A, dA ). 3. (A, d) is formal if and only if there is an isomorphism ϕ : (∧V, D) → (∧V, d) of the form ϕ = id + ϕ1 + ϕ2 + · · · with ϕq (Vp ) ⊂ (∧V)p−q . Definition 2.96 The Sullivan model (∧V, D) is called the filtered model of the cdga (A, dA ). We want to use Theorem 2.95 (3) to develop an obstruction theory for formality, so we note that it follows from a more general uniqueness result

2.7

Formality

for filtered models (see [132, Theorem 4.4]) and the fact that, if (A, dA ) is formal, then the bigraded model is a minimal model and a filtered model for (A, dA ). Now let’s see what necessary condition Theorem 2.95 (3) leads to. Suppose ϕ : (∧V, D) → (∧V, d) is an isomorphism with ϕ of the form ϕ = id + ϕ1 + ϕ2 + · · · . From the relations D2 = 0, ϕ(ab) = ϕ(a)ϕ(b) and dϕ = ϕD we obtain ⎧ 2 ⎨ d2 = 0 , ϕ (ab) = ϕ1 (a)b + aϕ1 (b) , ⎩ 1 dϕ1 = d2 + ϕ1 d . In particular, d2 and ϕ1 are derivations that satisfy ϕ1 (Vp ) ⊂ (∧V)p−1 and d2 (Vp ) ⊂ (∧V)p−2 . We are therefore motivated to define the graded Lie algebra of derivations Derq,s as the set of derivations θ of ∧V that decrease the lower degree by q and increase the usual degree by s: ⎧ ⎨ θ(V r ) ⊂ (∧V)r+s , p−q p ⎩ θ(ab) = θ(a) · b + (−1)s·|a| a · θ (b) . The commutator with d and the commutator bracket give Der∗,∗ the structure of a differential graded Lie algebra (Der∗,∗ , D). The differential D is given by

D(θ) = d ◦ θ − (−1)s θ ◦ d , so that (recalling d(Vp ) ⊂ (∧V)p−1 ) we have

D : Derq,s −→Derq+1,s+1 . Note that, for each θ ∈ Derq,s , q > 0, it is certainly true that θ q+1 (x) = 0 for x ∈ Vq . Therefore the formula eθ =

θn n≥0

n!

= id + θ +

θ2 + ··· 2

gives a well defined automorphism of ∧V. Now we can construct a sequence of obstructions to the formality of a cdga (A, dA ). Let (∧V, D) be a filtered model of (A, dA ). Since D2 = 0, we have dd2 + d2 d = 0 since it is the only part of D2 that decreases the lower degree by 3. The derivation d2 is thus a cycle in Der2,1 . If there is an isomorphism ϕ : (∧V, D) → (∧V, d) of the form ϕ = id + ϕ1 + · · · , then d2 = dϕ1 − ϕ1 d = D(ϕ1 ).

97

98

2 : Minimal models

Therefore the first obstruction to formality is the class of d2 , [d2 ] ∈ H2,1 (Der∗,∗ ) . On the other hand, if the class of d2 is zero, then there is a derivation ϕ1 such that d2 = dϕ1 − ϕ1 d. We then can form the automorphism e−ϕ1 and consider the new Sullivan model (∧V, D ) with D = eϕ1 De−ϕ1 . By construction, we have an isomorphism e−ϕ1 : (∧V, D ) → (∧V, D), but the main advantage of D is that D − d decreases the lower gradation by at least three. We can see this by writing D = (id + ϕ1 + . . .)(d + d2 + . . .)(id − ϕ1 + . . .) = d + (ϕ1 d − d ϕ1 + d2 ) + terms decreasing lower degree by at least 3 = d + terms decreasing lower degree by at least 3 since ϕ1 d − d ϕ1 + d2 = 0 by assumption. We can therefore write D = d + d3 + d4 + · · · with dq (Vp ) ⊂ (∧V)p−q . We then see that the second obstruction is the class of d3 in H3,1 (Der∗,∗ ). Of course, this process leads to an inductive construction. Therefore, suppose by induction that (A, dA ) admits a filtered model of the form (∧V, Dr ) with Dr = d + dr + dr+1 + · · · with r ≥ 3. The element dr is a cycle in Derr,1 , and defines the (r − 1)st obstruction [dr ] ∈ Hr,1 (Der∗,∗ ). Suppose we have an isomorphism ϕ (r) : (∧V, Dr ) → (∧V, d) of the form ϕ (r) = id + ϕ1 + ϕ2 + · · · . Then ϕ1 d = dϕ1 since d2 = 0. Thus, e−ϕ1 is an automorphism of (∧V, d) and we obtain the composition e−ϕ1 ◦ ϕ (r) = id+ϕ2 +. . .. Hence, we can suppose that ϕ1 = 0. By the same process, we can (r) suppose that ϕ2 = ϕ3 = . . . = ϕr−2 = 0; that is, ϕ (r) = id + ϕr−1 + · · · , and (r)

(r)

dr = dϕr−1 − ϕr−1 d. Hence, [dr ] must be equal to zero in H ∗ (Der∗,∗ , D). Conversely, if we suppose that [dr ] = 0, we can then replace (∧V, Dr ) (r)

(r)

by (∧V, Dr+1 ) with Dr+1 = eϕr−1 Dr e−ϕr−1 . As above, we also obtain an (r)

isomorphism e−ϕr−1 : (∧V, Dr+1 ) → (∧V, Dr ). If all the obstructions [di ] vanish, then we have a sequence of isomorphisms ···

/ (∧V, Dr+1 )

e

−ϕ

(r) r−1

/ (∧V, Dr )

/ ···

/ (∧V, D),

where Dr − d decreases the lower gradation by at least r, and where (r)

e−ϕr−1 − id decreases the gradation by at least r − 1. In other words, each

2.7

Formality

stage of the construction creates a cdga which is closer to (∧V, d) than the previous stage. Therefore, the composition (r)

(r−1)

· · · ◦ e−ϕr−1 ◦ e−ϕr−2 ◦ · · · ◦ e−ϕ1 is a well-defined isomorphism between (∧V, d) and (∧V, D). We therefore have the Theorem 2.97 Starting from the filtered model (∧V, D) for (A, dA ), there is a sequence of obstructions in H ∗ (Der∗,∗ , D), [d2 ], [d3 ], . . . , [dr ], . . . such that the following conditions are equivalent: • [d2 ] = [d3 ] = . . . = [dr ] = 0; • (∧V, D) ∼ = (∧V, D ) with (D − d)(Vq ) ⊂ (∧V)q−r .

In particular, all the obstructions vanish if and only if the cdga (A, dA ) is formal. The following proposition will significantly simplify future computations of the obstructions [dr ]. Proposition 2.98 Let (∧V, D) be the filtered model of a simply connected space X with finite Betti numbers such that H >n (X; Q) = 0. Suppose D = d + dq + dq+1 + · · · and suppose there exists a derivation θ defined on 2n, add generators to the minimal model (in degrees greater than or equal to 2n) to kill all cohomology above degree 2n. Show by Theorem 2.74 (3) that this is a contradiction. Exercise 2.5 Suppose X is a finite CW complex that is (r − 1)-connected with dim X ≤ 4r − 4 and H 2q+1 (X; Q) = 0 for q ≥ 0. Show that X is formal. (Note that X is not assumed to be a manifold.) Exercise 2.6 Characterize the compact homogeneous spaces that are formal in terms of their minimal models. (Hint: see Theorem 2.77). Exercise 2.7 Let f : M1 → M2 be a quasi-isomorphism of semifree A-modules. Show that there is a morphism of differential A-modules g : M2 → M1 such that g ◦ f and f ◦ g are homotopic to the identity.

103

3

Manifolds

A smooth compact manifold has many properties that make it distinct from an ordinary topological space. From the topological viewpoint, the existence of Poincaré duality in (co)homology is crucial to almost any result about the manifold. From the geometric viewpoint, the existence of a Riemannian metric allows the manifold to be studied using analytic techniques. In subsequent chapters, we shall see how these two points of view mix together to yield interesting results in both directions: topology applied to geometry and geometry applied to topology. In this chapter, we show how minimal models of manifolds reflect the special properties of manifolds. In particular, we see how Poincaré duality plays a huge role in almost all aspects of the rational homotopy of manifolds, but especially in the geometric realization of algebraic data and in the problem of determining whether a manifold is formal. Also, restricting manifolds to satisfy certain properties constrains their minimal models as well, and we will see this clearly in the case of nilmanifolds and biquotients. Now let’s describe exactly what is in the chapter. In Section 3.1, we study how manifolds are linked to minimal models through Poincaré duality. In particular, we state the Barge–Sullivan realization criteria which tell us which minimal models contain manifolds inside their rational homotopy types. We also describe a model for connected sums, understand their rational homotopy groups and show which simply connected 4-manifolds are rationally elliptic. We end this section by giving proofs of the theorems of Miller and Stasheff, which state two important criteria for formality, using the obstruction theory of Subsection 2.7.2. In Section 3.2, we consider a particular class of manifolds (which we will meet throughout the book) called nilmanifolds. The minimal model of a nilmanifold is very special indeed and this means that we can understand many general properties of nilmanifolds. The construction of models can easily be extended to the case of cdga’s equipped with the action of a finite group and we develop this in Section 3.3. This then leads to an understanding of the cohomology of the classifying

3.1

Minimal models and manifolds

space of a compact connected Lie group. For instance, we prove the well-known isomorphism H ∗ (BG; Q) ∼ = H ∗ (BT; Q)W(G) between the cohomology of BG and the invariant subalgebra of the cohomology of the classifying space of a maximal torus T of G under the action of the Weyl group W(G) of G. Section 3.4 prepares the way for one of the main objects of Chapter 6, the biquotients G//H of a compact connected Lie group G by a closed connected subgroup H. We define biquotients here and give their principal properties. We also show how Sullivan’s theory allows the study of biquotients through a special minimal model obtained from a description of G//H as a certain type of pullback. Finally, in Section 3.5, we describe the canonical model for a Riemannian manifold in terms of Hodge theory. While this type of model is not well-known, its potential for application in geometry is apparent from Theorem 3.54, which says that the real homotopy theory of an isometry of a closed simply connected Riemannian manifold is determined by the induced homomorphism on real cohomology.

3.1


In order to apply the algebra of minimal models to geometry, we have to understand how algebraic data can be realized geometrically. We have already seen in Subsection 2.6.1 that a minimal model may be realized by a space. While it is useful to have spatial realizations of models, for geometry it is essential to have realizations as closed manifolds. This, of course, leads to many restrictions on the original algebra. Let’s consider this now.

3.1.1

Sullivan–Barge classification

The most fundamental algebraic aspect of compact manifolds is the fact that their cohomology satisfies Poincaré duality. We want to be able to realize minimal models by manifolds, so we need to build in the Poincaré duality requirement. Definition 3.1 A cga H is a lk-Poincaré duality algebra of dimension n if each H q is of finite lk-dimension, H n = lkω, H >n = 0 and the multiplication induces a nondegenerate bilinear pairing H q ⊗ H n−q → H n ∼ = lk for 0 ≤ q ≤ n. Let (∧V, d) be a minimal model, with H 1 (∧V, d) = 0 and such that H ∗ (∧V, d) is a Poincaré duality algebra of dimension n. Also suppose that cohomology classes pi in degree 4i for 4i < n have been chosen. Let p denote the collection {pi }. A realization of the pair ((∧V, d), p) is a manifold

105

106

3 : Manifolds

whose minimal model is (∧V, d) and whose Pontryagin classes are the pi . The theorem of Sullivan and Barge (see [21], [246]) gives necessary and sufficient conditions for the realization of this data. Note first that in case the dimension is 4k, we have a quadratic form on H 2k whose signature is related to the Pontryagin numbers by the Hirzebruch signature formula (see [141]). Theorem 3.2 ([246]) Let (∧V, d) be a Sullivan model whose cohomology satisfies Poincaré duality with a fundamental class in dimension n and V 1 = 0. We also choose cohomology classes p = {pi } ∈ H 4∗ (∧V, d). 1. If n is not of the form 4k, then there is a compact simply connected manifold that realizes the pair ((∧V, d), p). 2. If n = 4k, and the signature is zero, there is a compact simply connected manifold that realizes the pair ((∧V, d), p) if and only if the quadratic form on H 2k is equivalent over Q to a quadratic form ±x2i . 3. If n = 4k and the signature is nonzero, then there exists a compact simply connected manifold realizing the pair ((∧V, d), p) if and onlyif the quadratic form on H 2k is equivalent over Q to a quadratic form ±x2i and the Pontryagin numbers are numbers satisfying the congruence of a cobordism [244]. In other words, the conditions that are necessary for the realization of algebraic data by a closed manifold are also sufficient. To see if a quadratic form has the form ±x2i over Q is not easy in general. Let’s look at an elementary example to illustrate this point. First, the rational quadratic form 2x2 + 2y2 can be written (x + y)2 + (x − y)2 , and therefore has the form desired. On the other hand, the quadratic form x2 + 2y2 cannot be written in the form ±z2 ± t 2 by a standard computation. So we have, in particular, that the graded algebra ∧(x, y)/(2x2 − y2 , xy), with |x| = |y| = 2, is not the cohomology of a manifold. The beginning reader will check that the quadratic form associated to this algebra is x2 + 2y2 .

3.1.2

The rational homotopy groups of a manifold

Let M be a simply connected n-dimensional compact manifold, p a point in M, and D an n-dimensional disk centered at p in M. The spaces M = M\{p} and M = M\ int D have the same homotopy type. Denote by ϕ : Sn−1 → M the inclusion of the boundary of D. Then M is homeomorphic to the space obtained by attaching an n-dimensional cell to M along the map ϕ. Theorem 3.3 ([87, Section 37]) Let M be a simply connected manifold such that the cohomology algebra requires at least two generators. Then,

3.1


1. the inclusion i : M → M induces a surjective map i# : π∗ (M ) ⊗ Q → π∗ (M) ⊗ Q ; 2. the kernel of i# is a free graded Lie algebra; 3. if we filter the Lie algebra π∗ (M ) ⊗ Q by the powers of the ideal generated by [ϕ], we obtain a filtered Lie algebra whose graded associated Lie algebra is the free product of Lie algebras L([ϕ]) (π∗ (M) ⊗ Q) ; 4. if ϕ is an indecomposable element in the Lie algebra π∗ (M ) ⊗ Q, then we have an isomorphism of graded Lie algebras π∗ (M ) ⊗ Q ∼ = L([ϕ]) (π∗ (M) ⊗ Q) . Corollary 3.4 (The rational homotopy of a connected sum) Suppose M and N are n-dimensional manifolds whose cohomologies require at least two generators. Then there is a filtration on π∗ ((M#N)) ⊗ Q such that the (M) ⊗ Q ) (π∗ (N) ⊗ associated graded Lie algebra is isomorphic to (π ∗ Q) L(x) for some element x in degree n − 1. In particular, we have an isomorphism of graded vector spaces L(x) . π∗ ((M#N)) ⊗ Q ∼ = (π∗ (M) ⊗ Q) (π∗ (N) ⊗ Q) Proof Denote by ψ1 and ψ2 the attaching maps of the top cells in M and N. By Theorem 3.3, the inclusion i : M ∨ N → M ∨ N induces a surjective map on the rational homotopy groups (see Example 2.58). The connected sum M#N is obtained from M ∨ N by attaching a cell along ψ1 + ψ2 . Therefore we have a factorization of i as j

q

M ∨ N → M#N → M ∨ N , where j is the canonical injection and q pinches the tube Sn−1 × [0, 1] connecting the two components to a point. This shows that the pinching map q induces a surjective map on the rational homotopy groups. Filtering π∗ ((M ∨N ))⊗ Q by the ideal generated by [ψ1 ] and [ψ2 ], we get a filtered Lie algebra whose associated graded Lie algebra is (π∗ (M) ⊗ Q) (π∗ (N) ⊗ Q) L([ψ1 ]) L([ψ2 ]) . by [ψ1 ],we get Therefore filtering π∗ ((M#N)) ⊗ Q by the ideal generated a graded Lie algebra isomorphic to (π∗ (M) ⊗ Q) (π∗ (N) ⊗ Q) L(x), with x = [ψ1 ].

107

108

3 : Manifolds

Remark 3.5 Note that, under the hypothesis of Corollary 3.4, the connected sum M#N is a rational hyperbolic space. Example 3.6 (Model of a connected sum) Let Mn and N n be simply connected compact n-manifolds. Denote by q the pinching map M#N → M∨N and by ϕ : MM∨N → MM#N the minimal model of q. Denote also by ωM ∈ MM∨N and ωN ∈ MM∨N cocycles representing the fundamental classes of M and N. Since [ϕ ∗ (ωM )] = [ϕ ∗ (ωN )] is the fundamental class of M#N, we can introduce a new generator x and define an extension of ϕ, ϕ : (MM∨N ⊗ ∧x, d) → MM#N , by putting dx = ωM − ωN . Since H ≤n (ϕ) is an isomorphism and we know that H >n (M#N; Q) = 0, to obtain a quasi-isomorphism, we have only to inductively introduce new generators xi in degrees ≥ n so that H >n (MM∨N ⊗ ∧(x, xi ), d) = 0. Recall that in Example 2.47 we have given a process to construct the minimal model of M ∨ N. That computation combined with the present process gives a procedure to derive a Sullivan model of M#N from the minimal models of M and N. When the algebras H ∗ (M; Q) and H ∗ (N; Q) are not generated by only one element, then the map π∗ (q) ⊗ Q is surjective, which implies that ϕ is injective. The relative minimal model we have constructed above is then the minimal model of M#N. Example 3.7 The minimal model of a connected sum M#N depends on the choice of the fundamental classes ωN and ωM of N and M. If we reverse the orientation of N the fundamental class becomes −ωN , and we have dx = ωM + ωN . This change of orientation can change the rational homotopy type. This is the case when M = N = CP(2). Write H ∗ (M; Q) = Q[x]/x3 and H ∗ (N; Q) = Q[y]/y3 . If the fundamental classes are x2 and y2 , the minimal model for M#N = CP(2)#CP(2) is (∧(x, y, z, t), d), dx = dy = 0, dt = xy, dz = x2 − y2 . By Exercise 2.4, we can see that no other generators are required. Hence, CP(2)#CP(2) is rationally elliptic. If we reverse the orientation in N, we get (∧(x, y, z , t ), d), dx = dy = 0, dt = xy, dz = x2 + y2 and this is a model of CP(2)#CP(2), the blow-up of CP(2) at a point (see Subsection 8.2.1). It is quite easy to see that there is no isomorphism between the two cdga’s (see Exercise 3.2). Therefore, the two manifolds do not have the same rational homotopy type. Example 3.8 (Rationally elliptic 4-manifolds) Rational homotopy conditions imposed on manifolds often restrict possibilities greatly. For instance, which rational homotopy types of closed simply connected 4-manifolds are

3.1


rationally elliptic? Of course, the key properties that we shall use are contained in Theorem 2.75. In particular, there is the dimension formula (6) as well as the refinements (7) and (8) of Theorem 2.75. In what follows, for each i, we shall refer to the i-th rational homotopy group simply by πi . The first thing to notice is that, by (8), 0 ≤ dim π5 ≤ 1 and 0 ≤ dim π7 ≤ 1 and they cannot be non-zero simultaneously. Moreover, again by (8), if dim π7 = 1 or dim π5 = 1, then π3 = 0. Similarly, by (7), if π4 = 0, then π2 = 0 and, if π4 = 0, then dim π2 ≤ 2. Finally, by (7), dim π4 = 0 or dim π4 = 1. This means we can work with each case separately. So suppose π4 = Q. Then the dimension formula (6) restricts the possibilities for homotopy. For instance, if π5 = Q, then the formula becomes 4 = 5 − 4 + 1 = 2, which is not true. Thus, in this case we cannot have a degree five generator. The other two cases are when π7 = Q and when π5 = 0 = π7 . The respective dimension formulas are 4 = 7 − 4 + 1, which is true, and 4 = 3 dim π3 − 4 + 1 = 3 dim π3 − 3, which is false since 4 is not divisible by 3. The true case gives a model with a degree 4 generator and a degree 7 generator. The differential is then forced since we need finite cohomology. The model is therefore (∧(x4 , y7 ), dy = x2 ) and we recognize this as a model for S4 . Now suppose π4 = 0. Then, since dim π2 ≤ 2, the reader can show that the only dimension formula possibilities are 4 = 5 − dim π2 and 4 = 3 dim π3 − dim π2 . The first, with dim π2 = 1, gives a model (∧(x2 , y5 ), dy = x3 ) and this is a model for CP(2). The other can only hold when dim π3 = 2 and dim π2 = 2. The possible models are: • (∧(x1 , x2 , y1 , y2 ), dy1 = x2 , dy2 = x2 ); a model for S2 × S2 . 1 2 • (∧(x1 , x2 , y1 , y2 ), dy1 = x1 x2 , dy2 = x2 − x2 ); a model for CP(2)#CP(2). 1 2 • (∧(x1 , x2 , y1 , y2 ), dy1 = x1 x2 , dy2 = x2 + x2 ); a model for CP(2)#CP(2). 1

2

Therefore, the rational homotopy types of simply connected rationally elliptic 4-manifolds are given by S4 , S2 × S2 , CP(2), CP(2)#CP(2) and CP(2)#CP(2). See Exercise 3.1 for the case of elliptic 5-manifolds.

3.1.3

Poincaré duality models

A Poincaré duality model for a compact simply connected n-dimensional manifold M is a cdga (A, d) that satisfies the following properties.

1. There are quasi-isomorphisms (A, d) ←− MM −→ APL (M). 2. Ap = 0 for p > n, A0 = Q, A1 = 0, each Aq is finite dimensional and An = Qω, for ω ∈ An . 3. The map ϕ : Ap → Hom(An−p , Q) given by ϕ(a)(b) = λ if ab = λω, is an isomorphism.

109

110

3 : Manifolds

When (A, d) is a Poincaré duality model, then its dual Hom((A, d), Q) is a free (A, d)-module of rank one generated by the homomorphism that takes the value 1 on ω. When dim V < ∞ and V = V odd , then (∧V, d) is a Poincaré duality algebra. A result of Lambrechts and Stanley generalizes this fact (see [169]). Theorem 3.9 Every compact simply connected manifold admits a Poincaré duality model.

3.1.4

Formality of manifolds

Since we wish to study aspects of the geometry of manifolds using algebraic models, our first step might be to see if the particular manifold of interest is formal (see Section 2.7). Of course, there are some important classes of manifolds that are formal. For instance, spheres, Lie groups and Kähler manifolds (see Theorem 4.43) are formal spaces. The first important result on the formality of manifolds is due to Miller [192]. We give here a proof that is inspired by Miller’s, but which uses differential graded algebra models instead of Lie models. Our proof makes use of the obstruction theory for formality presented in Subsection 2.7.2. A very different and interesting proof can also be found in [97]. After Miller’s theorem, we give a powerful theorem of Stasheff [241] that says that an n-dimensional manifold M is formal if its (n − 1)-skeleton is. The original complicated proof of this result also used Lie models (which are generally good for situations where cell-attaching occurs), but here we stay within the world of differential graded algebras. Let’s begin now with Miller. Proposition 3.10 ([192]) Let M be a (p − 1)-connected compact manifold, p ≥ 2, of dimension m ≤ 4p − 2. Then M is formal. Proof To prove the theorem we will use the obstruction theory developed in Proposition 2.98. Let (∧V, d) denote the bigraded model of H = H ∗ (M; Q). q q By the connectivity hypothesis, V0 = 0 for q < p, and this implies Vr = 0 q for q < (r+1)p−r. By Poincaré duality, we have H = 0 for m−p < q < m and H m = Q · ω ∼ = Q. Before we look at the obstructions we make two observations. First of all, we notice that ∧+ V≥1 does not contain any non-zero cocycle. To see this, suppose that ∧+ V≥1 has nonzero cocycles. We denote by (xr ) an ordered basis of V≥1 such that d(xr ) ∈ ∧V0 ⊗ ∧(x1 , . . . , xr−1 ), and we denote by a a nonzero cocycle in ∧(x1 , . . . , xr ), where r is the smallest possible for a cocycle. We write a = an xnr + . . . a1 xr + a0 with

3.1


an = 0 and ai ∈ ∧(x1 , . . . , xr−1 ). Then d(an ) = 0 and, by our minimality condition, an is a constant. From da = 0, we then deduce that n an d(xr ) + d(an−1 ) = 0. This shows that by making a change of generators we can suppose dxr = 0, which is impossible. Therefore, no cocycles exist in ∧+ V≥1 . Poincaré duality is crucial in the proof. We will explicitly use the following property of Poincaré duality algebras: every linear map f : W ⊗ H m−r → H m factors as a composition ϕ⊗1

W ⊗ H m−r −→ H r ⊗ H m−r −→ H m . mult

Given f , for every element w ∈ W, we obtain by adjunction a linear map fw : H m−r → H m defined by fw (x) = f (w ⊗ x). By Poincaré duality, there is then an element xw ∈ H r such that fw (y) = xw · y for any y. We define ϕ(w) = xw linearly and obtain a map ϕ : W → H r such that f (w ⊗ x) = ϕ(w) · x. Now we can consider the obstructions to the formality. By Proposition 2.98, the only obstructions are [d2 : V2m−1 → (∧V0 )m ] and [d3 : V3m−1 → (∧V0 )m ] . The other obstructions are zero because Vq 1, da = 0 implies that p(dxr ) up +dup−1 = 0 as well. The element p xr up + up−1 is therefore a cocycle in (∧V)q for some q, 0 < q < n, and is a coboundary in (∧V)≤n : p xr up + up−1 = dv .

p−1 We then replace a by a = a − d 1p xr v and we obtain a new representative of α of the form a = xr

p−1

u p−1 + . . . + u 0 .

In conclusion, we can always suppose p = 1 and that a representative of α has the form a = u1 xr + u0 , with u1 ∈ ∧V0 , [u1 ] = 0, and xr ∈ Vn . We can now state and prove Stasheff’s theorem. Theorem 3.12 Let Mm be a simply connected compact manifold. If M\{∗} is formal, then M is formal. Proof Let (∧V, d) be the bigraded model of H = H ∗ (M; Q) with associated filtered model (∧V, D). Because M\{∗} is formal, we can assume

113

114

3 : Manifolds

that D = d on V 2, and let M#N be their connected sum. Suppose π1 (M) = 0 and |π1 (N)| ≥ 4. Let x ∈ π1 (M), x = 1, and let y1 , y2 , z = 1 ∈ π1 (N) be distinct nontrivial elements. Then the elements (xyi xyj )n xz belong to different conjugacy classes in π1 (M) ∗ π1 (N). This shows that π1 (M#N) = π1 (M) ∗ π1 (N) contains infinitely many conjugacy classes. It follows from Hadamard’s result that M#N admits infinitely many geometrically distinct closed geodesics. 1

Let us denote by LM = MS the space of continuous maps from S1 to M with the compact open topology and by H1 (S1 , M) the Hilbert manifold consisting of absolutely continuous maps from S1 into M with square integrable derivative. By a result of Klingenberg [154] the inclusion i : H1 (S1 , M) → LM is a homotopy equivalence (see [120]). Because of this equivalence, we shall abandon the notation H1 (S1 , M) and simply use LM even in situations where we obviously are referring to H1 (S1 , M). We can therefore consider the energy of a closed curve, E : LM → R, defined by 1 1 c (t), c (t) dt . E(c(t)) = 2 0 Recall now that if W is a Hilbert manifold, and f : W → R is a smooth map, then m ∈ W is a critical point if ∇(f )(m) = 0. In this case, ∇(E) = 0 produces the Euler–Lagrange equations for the energy functional E. Of course, these Euler–Lagrange equations are simply the geodesic equations (see [216, page 441]), so we have the Theorem 5.7 ([154]) The critical points of the energy function of a manifold are the constant maps and the closed geodesics. In order to appreciate the significance of this result, let’s see why every compact manifold must have at least one closed geodesic. The approach we describe is due to the combined efforts of many mathematicians including Birkhoff, Morse and Bott. Define Pn (M) = {(x1 , . . . , xn ) ∈ Mn | d(x1 , x2 )2 + · · · + d(xn , x1 )2 ≤ } where d(x, y) denotes the distance from x to y on M and 0 ≤ ≤ (M). Here, (M) is the particular number (called the injectivity radius) associated to M such that, for any x and y in M with d(x, y) ≤ (M), there exists a unique minimal geodesic joining x and y. Such an (M) always exists for a compact manifold. Clearly, Pn (M) is compact and every point of Pn (M) represents a closed geodesic n-gon with corners at the xi . We have an inclusion j : Pn (M) → LM (after parametrizing the n-gons with respect to arclength). Moreover, a theorem of Bott says that, for fixed r and all k ≤ r,

5.1

The closed geodesic problem

there exists nr such that, for n ≥ nr , ∼ =

j# : πk (Pn (M)) → πk (LM). From the point of view of homotopy theory then, the Pn (M) approximate LM better and better as n goes to infinity. Therefore, for homotopy questions, we have an effective finite dimensional reduction. Now let’s look at the situation from the viewpoint of critical point theory. We can define a type of energy function E : Pn (M) → R by E(x1 , . . . , xn ) =

n

d(xi , xi+1 )2

i=1

where xn+1 = x1 . This discretizes the energy functional E described above. It can be shown that a critical point of E gives rise to a polygon without corners with sides of equal length; that is, a closed geodesic. Note that the section s : M → Pn (M) ⊂ LM embeds M as the “trivial” critical points which are absolute minima for the energy functional E having E = 0. Of course, we want to find critical points besides these trivial ones and this is the content of the theorem of Lusternik and Fet. Theorem 5.8 ([181]) On any compact simply connected smooth manifold, there exists a nontrivial closed geodesic. Proof Suppose no critical point of E exists with E > 0. By [66, Theorem 1.17], there is a deformation of Pn (M) into an open neighborhood of M which may be taken as close to M as we desire. In particular, Pn (M) deforms into a tubular neighborhood of M, which itself deformation retracts onto M. If H is the total deformation of Pn (M) into M and h = H1 , then we have s ◦ h idPn (M) , where s is the inclusion M → Pn (M). In particular, this implies that h# : π∗ (Pn (M)) → π∗ (M) is injective. Now suppose the first nonzero homotopy group of M occurs in degree r. Because πk (M) = πk+1 (M), the splitting πk (LM) ∼ = πk (M) ⊕ πk (M) says that the first nonzero homotopy group of LM occurs in degree r − 1. If we choose n large enough, then this will also be true for Pn (M). Thus, we may assume that πt (Pn (M)) = 0 for t < r − 1 and πr−1 (Pn (M)) = 0. The injectivity of h# then implies that πr−1 (M) = 0 and this is a contradiction to the assumption that the first nonzero homotopy group occurs in degree r. Hence, our original supposition is incorrect and a non-trivial critical point of E exists in Pn (M). Thus, there exists a nonconstant closed geodesic on M. Of course, it is a long way from proving the existence of a single closed geodesic to showing that a manifold admits an infinite number of closed

209

210

5 : Geodesics

geodesics. The crucial step was taken by Gromoll and Meyer in 1969, so let’s turn to that now. The circle acts on LM by rotation along the curves A : S1 × LM → LM

A(t, c)(s) = c(t + s mod Z) .

The orbit of a curve, S1 (c), is a submanifold of LM. Clearly, two geodesics c and c are geometrically distinct if c ∈ S1 (c). Note that if c is a critical point of E, then the same is true for all the points in S1 (c). Relations between critical points of the energy function and the homology of LM can then be established by adapting Morse theory to the infinite dimensional case. This gives the following theorem due to Gromoll and Meyer. Theorem 5.9 ([117]) Let M be a compact simply connected manifold. If the sequence of Betti numbers of LM is not bounded, then M admits infinitely many geometrically distinct closed geodesics. And this then leads to the main question in the area for which algebraic models are relevant. Question 5.10 When is the sequence of Betti numbers of the free loop space an unbounded sequence? This question will be answered in Proposition 5.14.

5.2

A model for the free loop space

The Gromoll–Meyer theorem reduces the closed geodesic problem to the problem of the computation of the cohomology of the free loop space. In this section we give a procedure to compute the minimal model for the free loop space LX of a simply connected space X from its minimal model (∧V, d). Theorem 5.11 Let X be a simply connected space with minimal model (∧V, d). Then a model for the free loop space, LX, is given by (∧V ⊗ ∧sV, δ) with δ(v) = dv and δ(sv) = −sd(v) where s is the derivation defined by s(v) = sv.

5.2

A model for the free loop space

Proof The space LX is the pullback in the following diagram of fibrations / X [0,1]

LX p0

X

(p0 ,p1 )

/ X×X

where pj (c) = c(j). Denote by i : X → X [0,1] the map that associates to a point x the constant path at x. The map i is a homotopy equivalence making the following diagram commutative i

/ X [0,1] X E EE EE EE (p0 ,p1 ) EE " X×X

This implies that (p0 , p1 ) and have the same relative minimal model. Denote by (∧V1 , d) and (∧V2 , d) two copies of (∧V, d). By Example 2.48, a minimal model of is given by the multiplication µ : (∧V1 , d) ⊗ (∧V2 , d) → (∧V, d) . By Example 2.48 again, a relative minimal model of µ is given by the following commutative diagram µ

/ (∧V, d) (∧V1 , d) ⊗ (∧V2 , d) TTTT O TTTT TTTT ϕ TTTT * (∧V1 ⊗ ∧V2 ⊗ ∧sV, D)

where ϕ is a quasi-isomorphism with ϕ(sv) = 0, and where D(sv) − v2 + v1 is a decomposable element in ∧V1 ⊗ ∧V2 ⊗ ∧sV (for, recall that a decomposable element in ∧T is an element in ∧≥2 T). By Theorem 2.70 together with the standard pullback description of LX, we see that a model for LX is given by the tensor product (∧V, d) ⊗(∧V1 ⊗∧V2 ) (∧V1 ⊗ ∧V2 ⊗ ∧sV, D) . The proof of the theorem will follow from the construction of an isomorphism (∧V, d) ⊗(∧V1 ⊗∧V2 ) (∧V1 ⊗ ∧V2 ⊗ ∧sV, D) → (∧V ⊗ ∧sV, δ) .

211

212

5 : Geodesics

For this purpose, we give a more explicit construction of the cochain algebra ˆ D) constructed (∧V1 ⊗∧V2 ⊗∧sV, D). Consider the model (∧V ⊗∧V¯ ⊗∧V, in Chapter 2 for the definition of left homotopies (Definition 2.18) with differential D(v) = dv , D(ˆv) = 0 , D(¯v) = vˆ , and V¯ = sV. This model is equipped with a derivation s satisfying s(v) = v¯ , s(¯v) = 0 , s(ˆv) = 0 . The commutator θ = sD + Ds is a derivation of degree 0 and the morphism ˆ D) eθ : (∧V, d) → (∧V ⊗ ∧V¯ ⊗ ∧V, is a well-defined morphism of cdga’s. By definition, eθ (v) = v + vˆ + decomposable . We then consider the isomorphism of algebras ψ : ∧ V1 ⊗ ∧V2 ⊗ ∧sV −→ ∧V ⊗ ∧V¯ ⊗ ∧Vˆ , defined by ψ(v1 ) = v, ψ(v2 ) = eθ (v), and ψ(sv) = v¯ . We make ψ an isomorphism of cdga’s by defining D on ∧V1 ⊗ ∧V2 ⊗ ∧sV by the formula D = ψ −1 Dψ . This differential D then inherits the good properties of D on ∧V1 ⊗ ∧V2 ⊗ ∧sV; that is, D = d on V1 and on V2 and D(sv) − v2 + v1 is a decomposable element. We now define a morphism of cdga’s ˆ D) → (∧V ⊗ ∧sV, δ) q : (∧V ⊗ ∧V¯ ⊗ ∧V, by q(v) = v, q(¯v) = sv and q(ˆv) = −sd(v). A simple computation shows that q(sD + Ds) = 0. We then form the composition q ◦ ψ : (∧V1 ⊗ ∧V2 ⊗ ∧sV, D) → (∧V ⊗ ∧sV, δ). By the observation that q(sD + Ds) = 0, we have q ◦ ψ(v1 ) = v, q ◦ ψ(v2 ) = v and q ◦ ψ(sv) = sv. This induces the promised isomorphism 1⊗(q◦ψ)

(∧V, d) ⊗(∧V1 ⊗∧V2 ) (∧V1 ⊗ ∧V2 ⊗ ∧sV, D) −−−−−−−→(∧V ⊗ ∧sV, δ).

For the sake of simplicity, in the minimal model of the free loop space, (∧V ⊗ ∧sV, δ), we will write x¯ instead of sx for an element of sV.

5.3

A solution to the closed geodesic problem

Example 5.12 Let M be the sphere S2 . A minimal model for S2 has the form (∧(x, y), d) with |x| = 2, |y| = 3, dx = 0 and dy = x2 . A minimal model for LM is thus given by the cochain algebra (∧(x, y, x, ¯ y¯ ), δ), with |x| ¯ = 1, |¯y| = 2, δx = 0, δy = x2 , δ(x) ¯ = 0, δ(¯y) = −2xx. ¯ We can compute the cohomology to be ¯ x, x¯ ¯ y, x¯y − 2xy, ¯ x¯ ¯ y2 , . . .. H ∗ (LM; Q) = x, Note that dim H i (LM; Q) = 1 for all i and that all cup products are zero. Nevertheless, the free loop space of S2 is not formal because there are many Massey products such as x¯y − 2xy. ¯

5.3

A solution to the closed geodesic problem

In the historical development of rational homotopy theory, one of the first major applications of the theory was the theorem of Vigué-Poirrier and Sullivan on closed geodesics [261]. Theorem 5.13 If M is a compact simply connected Riemannian manifold whose rational cohomology algebra requires at least two generators, then M has infinitely many geometrically distinct closed geodesics. This theorem is in fact a direct consequence of Theorem 5.14 below combined with Theorem 5.9. Proposition 5.14 ([261]) Let M be a simply connected space whose rational cohomology is finite dimensional. The following conditions are equivalent. 1. The sequence of rational Betti numbers of LM is unbounded. 2. The cohomology algebra H ∗ (M; Q) requires at least two generators. 3. The dimension of πodd (M) ⊗ Q is at least two. Proof If the cohomology H ∗ (M) is singly generated, then either H ∗ (M) = ∧x, with x in odd degree, or else H ∗ (M) = Q[y]/yn . In the first case, the minimal models of M and LM are (∧x, 0) and (∧(x, x), ¯ 0). The Betti numbers of LM are either 0 or 1, so they are bounded. In the second case, the minimal model of M is (∧(y, z), d) with dy = 0 and dz = yn . The minimal model for LM is thus (∧(y, z, y¯ , z¯ ), δ) with y¯ in odd degree, z¯ in even degree, δ(¯y) = 0 and δ(¯z) = −nyn−1 y¯ . Since the ideal generated by z and yn is acyclic, the complex is quasi-isomorphic to the quotient complex

¯ z) = −nyn−1 y¯ . δ(¯ ∧(y, y¯ )/yn ⊗ ∧¯z, δ¯ ,

213

214

5 : Geodesics

A standard computation shows that the reduced cohomology of LM is therefore ∧+ (y, y¯ )/(yn , y¯ yn−1 ) ⊗ ∧¯z . Once again the Betti numbers are bounded. Observe also that in both cases, the dimension of πodd (M) ⊗ Q is one. This shows (1) ⇒ (2) and (3) ⇒ (2). Suppose now that the cohomology requires at least two generators. We first prove that the dimension of πodd (M) ⊗ Q is at least two, i.e. (2) ⇒ (3). If there is no odd generator, the differential is zero for degree reasons, and the cohomology is infinite, which is impossible. Suppose there is only one odd generator y. If dy = 0, then the first even generator is a cocycle, and no power of it is a coboundary because there is no other odd generator, and so the cohomology is infinite, which is impossible. Therefore dy is nonzero. We now order the generators by degree and we obtain (∧(x1 , . . . , xn , y, xn+1 , . . .), d) , with |xi | even. Since dy is not a zero divisor in ∧(x1 , . . . , xn ), the cohomology of (∧(x1 , . . . , xn , y), d) is ∧(x1 , . . . , xn )/(dy) and therefore, the global cohomology is ∧(x1 , . . . , xn )/(dy) ⊗ ∧(xn+1 , . . .) . If n ≥ 2 the dimension of ∧(x1 , . . . , xn )/(dy) is infinite and the cohomology is generated by the xi . If n = 1, we must have another generator x2 of even degree because the cohomology requires at least two generators. Now d(x2 ) = 0 because there is no cocycle in Qy · ∧(x1 ). It follows that the cohomology is infinite. This proves that we have at least two odd generators. We finish by proving (2) ⇒ (1). By our discussion above, we can assume that the model for M contains at least two odd generators. Therefore, let y1 and y2 be the first two odd generators and denote the generators by increasing degrees: x1 , . . . , xn , y1 , xn+1 , . . . , xr , y2 , . . . The differential d is given on the first generators by polynomials P1 , Qi and P2 : dx1 = 0, dx2 = 0, . . . , dxn = 0 , dy1 = P1 (x1 , . . . xn ) , dxn+1 = y1 Q1 (x1 , . . . , xn ), . . . , dxr = y1 Qr (x1 , . . . , xr−1 ), dy2 = P2 (x1 , . . . , xr ) The ideal J generated by x1 , . . . xr is then a differential ideal in the minimal model of the free loop space (∧V ⊗ ∧sV, δ). The quotient (A, δ) = (∧V ⊗ ¯ δ), where V¯ = ∧sV, δ)/J is the minimal cdga (A, δ) = (∧(y1 , y2 , . . .) ⊗ ∧V, sV as usual. In this quotient, the elements x1 · · · xr y1 p y2 q , p, q ≥ 1

5.4

A-invariant closed geodesics

are cocycles that induce linearly independent classes in cohomology. It follows that the Betti numbers of the cohomology of (A, δ) are unbounded (see the loop space analogue in Exercise 5.5). We now apply Lemma 5.16 inductively (beginning with xr ) to deduce that the Betti numbers of (∧V ⊗∧sV, δ) are also unbounded. Lemma 5.15 Let (A, d) be a connected cdga, y an odd dimensional generator, and (A, d) → (A ⊗ ∧y, d) a relative minimal model. If the Betti numbers of (A ⊗ ∧y, d) are unbounded, then the same is true for (A, d). Proof We filter the cdga A ⊗ ∧y by the ideals A≥p ⊗ ∧y. The E2 -term of the associated spectral sequence is H ∗ (A, d) ⊗ ∧y. Since the spectral sequence converges to the cohomology of (A ⊗ ∧y, d), we deduce that dim H q (A ⊗ ∧y, d) ≤ dim H q (A, d) + dim H q−m (A, d) , where m denotes the degree of y.

Lemma 5.16 Let (∧V, d) be a minimal model and x be a closed even dimensional generator of ∧V. If the Betti numbers of the quotient cdga ¯ are unbounded, then the Betti numbers of (∧V, d) are also (∧V/(x), d) unbounded. Proof The quotient is quasi-isomorphic to the Sullivan model (∧V ⊗ ∧y, d) where dy = x. The result follows then directly from Lemma 5.15. Example 5.17 Note also that S2 does not fall under the hypotheses of Proposition 5.14. However, as we mentioned in the introduction, the combined work of V. Bangert [19], J. Franks [99] and N. Hingston [140] led to a proof of the closed geodesic problem for S2 with any metric.

5.4 A-invariant closed geodesics The symmetry of a Riemannian manifold is reflected by its isometry group. Therefore, we can try to mix together the geometric content engendered by knowledge of the geodesics on a manifold and the symmetry information underlying the existence of an isometry. To see how to accomplish this, let A denote an isometry of a compact simply connected Riemannian manifold M. Definition 5.18 A geodesic γ (t) is called A-invariant if there exists some T ∈ R such that γ (t + T) = A(γ (t)) for all t ∈ R. Example 5.19 Let M be the sphere S2 with the usual metric. Denote by A the antipodal map; then the great circles are A-invariant geodesics. When

215

216

5 : Geodesics

M is a flat torus and A is the involution A(x, y) = (y, x) then the A-invariant geodesics are the lines in the square [0, 1]2 that are parallel to the ascending diagonal. Let’s denote the interval [0, 1] by I. The natural space to consider for the I of paths c : I → M satisfying study of A-invariant geodesics is the space MA I c(1) = A(c(0)). Observe that MA is the pullback / MI

I MA

M

(id,A)

(p0 ,p1 )

/ M×M

and note that, if A = id, we recover the free loop space LM. Just as for loop spaces, the space of paths MI has the same homotopy type as the Hilbert manifold H1 (I, M) consisting of absolutely continuous curves σ : I → M with square integrable derivative. We can then take the pullback of the diagram 1 (I, M) HA p

G(A)

/ H1 (I, M)

p

/ M×M

where p(c) = (c(0), c(1)) and G(A) ⊂ M × M is the graph of the isometry A. Observe that G(A) is homeomorphic to M. Since p is a submersion, the 1 (I, M) has the homotopy type of MI . Hilbert submanifold HA A In [120], K. Grove extended the classical theory of closed geodesics to the A-equivariant situation and proved the following. Theorem 5.20 ([120]) The critical points of the energy function 1 (I, M) → R are the closed A-invariant geodesics and the constant maps HA at fixed points. Note that the set of fixed points for A, Fix A, is a disjoint union of totally geodesic submanifolds because, when A(x) = x, A∗ is the identity when restricted to the tangent space of the submanifold, and expx ◦A∗x = A◦expx . The starting point of the study of A-invariant geodesics is the strong relation I: between Fix A and MA Theorem 5.21 ([120]) If Fix A = ∅, then M has an A-invariant geodesic. If Fix A = ∅, and M has no nontrivial A-invariant geodesic, then the natural I is a homotopy equivalence. injection Fix A → MA

5.4


We use Theorem 5.21 to obtain a link between the existence of Ainvariant geodesics on M and algebraic invariants of M such as homotopy groups or the Euler characteristic. In order to obtain explicit criteria for the existence of A-invariant geodesics, we must study certain properties of the fibration I M → MA → G(A) . If we replace A by an isometry that is homotopic to it, then the homotopy type of the fibration does not change. This is interesting because the isometries form a compact Lie group whose elements of finite order are dense [205]. Therefore, A is homotopic to an element of finite order. We can thus suppose without loss of generality that A has finite order, say Ak = 1, for some integer k. ∼ =

Let’s first recall the isomorphism θ : πq (M) → πq+1 (M). Denote by x0 the base point of M, x¯ 0 the constant loop at the base point and PM the subspace of MI consisting of paths c starting at x0 . Each continuous map c : (I q , ∂I q ) → (M, x0 ) lifts into a map c˜ : I q → PM defined by c˜ (t1 , · · · , tq )(t) = c(t1 , · · · , tq−1 , ttq ). By restriction to I q−1 × {1}, this gives the isomorphism θ : πq (M) → πq−1 (M). The inverse map θ −1 is defined by θ −1 (c)(t1 , · · · , tq ) = c(t1 , · · · , tq−1 )(tq ) . p

Now, in the fibration M → MI → M × M, p(c) = (c(0), c(1)), each continuous map c = (c1 , c2 ) : (I q , ∂I q ) → (M × M, (x0 , x0 )) lifts into the map c1 · c2 : I q → MI , where · denotes composition of paths and c¯ 1 the path inverse to c1 . Let δM denote the connecting map in the homotopy long exact sequence for the fibration p, δM (c) = c1 · c2 restricted to I q−1 × {1}. This implies that the composition θ −1

δM

πq (M) ⊕ πq (M) −→ πq−1 (M) −→ πq (M) maps (a, b) to b − a when q > 1 and to b · a−1 when q = 1. Now denote by ϕ : M → G(A) the homeomorphism defined by ϕ(m) = (m, A(m)). We then have: Lemma 5.22 Let δ : πq (G(A)) → πq−1 (M) be the connecting map of the I → G(A). Then the composition fibration M → MA πq (ϕ)

δ

θ −1

ρ : πq (M) −→ πq (G(A)) −→ πq−1 (M) −→ πq (M)

217

218

5 : Geodesics

satisfies ρ(x) = π1 (A)(x) · x−1 for q = 1 and ρ(x) = πq (A)(x) − x when q ≥ 2. Proof The proof follows directly from the above computations and the commutativity of the following pullback diagram (where G(A) → M × M is the natural inclusion) M

=

/ M

I MA

/ PM

p

G(A)

/ M×M

Theorem 5.23 ([121]) Let M be a compact connected Riemannian manifold. Then M has a nontrivial A-invariant geodesic if one of the following is satisfied. I ) = 0 for some q ≥ 1. πq (MA πq (A) − id is not an isomorphism for some q ≥ 2. π1 (A)(x) = x for some x ∈ π1 (M). rank πk (M) is odd for some k ≥ 2. m π1 (M) is finite and Ap is homotopic to the identity for some prime p and some integer m. 6. Fix A is not a connected set and M is simply connected.

1. 2. 3. 4. 5.

Proof Suppose there is no nontrivial A-invariant geodesic. Then, by Theorem 5.21, Fix A = ∅. Furthermore, Fix A has no component of dimension at least one because, otherwise, this component would contain a nontrivial closed geodesic – that is, an A-invariant geodesic. Therefore, I. since M is compact, Fix A is a finite set that has the homotopy type of MA I In particular, the homotopy groups πq (MA ) = 0 for q ≥ 1 and the connecting map is an isomorphism for q ≥ 1. By the description of the connecting map in Lemma 5.22, this proves the assertions (1), (2) and (3). (4) follows from the fact that, if B is an automorphism of finite order of some Zm , with m odd, then id − B is not an isomorphism. In fact, B is represented by a matrix with integer coefficients, which means that the roots of the characteristic polynomial are algebraic integers. On the other hand, since a power of B, say k, is the identity, the eigenvalues are kth roots of unity. Moreover, since the matrix has integer coefficients, if eiθj is

5.4


an eigenvalue, then its conjugate e−iθj is too. Since m is odd, at least one eigenvalue is a real number. If 1 is an eigenvalue, then clearly id − B is not an automorphism. So suppose 5 that −1 is an eigenvalue and id − B is an isomorphism. The product (1 − eiθj )(1 − e−iθj ) is an algebraic integer since it is a product of algebraic integers. On the other hand, this product is equal to det(id − B)/2 and, therefore, ±1/2 must be an algebraic integer. But this is impossible because a rational number that is an algebraic integer must be an integer. (5) The universal covering is a compact manifold and therefore for some q ≥ 2 the rank of πq (M) is nonzero. Consider the morphism id − πq (A) on πq (M)/torsion. Recall that, over the field Fp with p elements, we have (x−y)p = xp −yp . Therefore since πq (M) is abelian and p is a prime number, we have m

m

(id − πq (A))p = id − (πq (A))p + pB m

for some B. Since (πq (A))p = id, we see that id − πq (A) is not an isomorphism on πq (M)/torsion. Therefore, id − πq (A) is not an isomorphism at all and we now use (2). I is connected and, therefore, (6) Since M is simply connected, the space MA cannot have the homotopy type of Fix A. Grove and Halperin used this result to obtain an existence theorem for A-invariant geodesics on a compact rationally elliptic manifold. Theorem 5.24 ([122]) Let M be a compact simply connected rationally elliptic Riemannian manifold. If M has no nontrivial A-invariant geodesic, then 1. The dimension n is even, and, if χ(M) = 0, n is congruent to 0 modulo 4. 2. The Euler–Poincaré characteristic is the square of another integer. Proof (1) Let ri = rank πi (M). Recall that the dimension n of the manifold is a formal consequence of the ranks of the homotopy groups (see Theorem 2.75): (2q − 1) · (r2q−1 − r2q ) . n= q

Since, by (4) of Theorem 5.23, all the ranks are even, we see that n is also even. Moreover when χ(M) = 0, then χ (M) = 0, so π q r2q−1 (M) = r (M) and therefore n is congruent to 0 modulo 4. q 2q (2) When χ (M) = 0, then χ(M) is a square. We can thus suppose that χ(M) > 0, so that χπ (M) = 0 (Theorem 2.75). Recall from Theorem 2.79 that the Poincaré series of M and the ranks of its homotopy groups are

219

220

5 : Geodesics

related by

dim Hq (M; Q) t q =

q

(1 − t 2i )r(2i−1) −r2i . i

1 − t 2i

Decompose in the form = (1 + t)(1 − t + t 2 − · · · − t 2i−1 ). Then, since i r2i−1 − r2i = 0, we have 5

i (1 − t

1 − t 2i

2i )r2i−1 −r2i

= =

5

i (1 + t)

5

i (1 − t

r2i−1 −r2i

5

i (1 − t

+ · · · − t 2i−1 )r2i−1 −r2i

+ · · · − t 2i−1 )r2i−1 −r2i .

Therefore taking the value at −1, we obtain χ(M) = (2i)r2i−1 −r2i . i

Since the ranks rj are even, we deduce that χ(M) is a square.

Corollary 5.25 Let A be an isometry of a compact simply connected rationally elliptic Riemannian manifold M. If either the dimension n is odd or the Euler–Poincaré characteristic is not a square, then M admits a nontrivial A-invariant geodesic. There are other manifolds that have the property that for any Riemannian metric, every isometry admits a nontrivial invariant geodesic. For instance Papadima has proved that this is the case for the homogeneous space G/K when K is a maximal torus (see [221, Theorem 1.4(i)]), or G is simple (see [222, Corollary 1.4(i)]). I because It is quite easy to compute a minimal model for the space MA I MA is the pullback of the following diagram of fibrations / MI

I MA

M

(1,A)

(p0 ,p1 )

/ M×M

Example 5.26 When M is the sphere S2 and A the antipodal isometry, then A induces multiplication by −1 on π2 (S2 ), so a model for (1, A) is given by ϕ : (∧(x, y, x , y ), d) → (∧(x, y), d) with d(x) = d(x ) = 0, d(y) = x2 , d(y ) = x 2 , ϕ(x) = x, ϕ(x ) = −x and I is thus obtained by taking the pushout ϕ(y) = ϕ(y ) = y. A model for MA

5.4


in the following diagram (see Theorem 2.70) (∧(x, y, x , y , x, ¯ y¯ ), d) O j

(∧(x, y), d) o

ϕ

(∧(x, y, x , y ), d)

x. In this diagram j is the inclusion and d(x) ¯ = x−x , d(¯y) = y−y −xx−x ¯ ¯ I It follows that a model (not minimal) for MA is given by

(∧(x, y, x, ¯ y¯ ), D) , D(x) ¯ = 2x , D(¯y) = 0 . This space has the rational homotopy type of S3 × S3 , and is very different from LS2 . For instance, H 1 (LS2 ; Q) is non-zero, but H 1 ((S2 )IA ; Q) = 0. Since we can take the isometry A to be of finite order, the cyclic group G generated by A acts on M. Therefore, Theorem 3.26 implies that the manifold M admits a G-equivariant minimal model (∧V, d). This means that A acts on V and that the action is extended in a multiplicative way to ∧V. In particular, V decomposes as V A ⊕ J where V A is the subvector space generated by the vectors that are invariant under A, and J is the sum of the eigenspaces corresponding to the eigenvalues different from 1. Since the ¯ denote model is equivariant, we also have d(J) ⊂ ∧V A ⊗ ∧+ J. Let (∧V A , d) the quotient of (∧V, d) by the differential ideal generated by J. Therefore the ideal I generated by J and sJ in the model of the free loop space (∧V⊗∧sV, δ) ¯ is a differential ideal, and we can form the quotient cdga (∧V A ⊗∧s(V A ), δ), ¯ ¯ where δ(sx) = −sd(x). ¯ ¯ where δ(sx) ¯ Lemma 5.27 The cdga (∧V A ⊗ ∧s(V A ), δ), = −sd(x) is the I minimal model for MA . Proof Following the steps in the construction of the minimal model for the free loop space given in Section 5.2, we give a natural G structure to ˆ D), by g · v¯ = g · v and g · vˆ = g6 (∧V ⊗ ∧V¯ ⊗ ∧V, · v. The isomorphism of ¯ algebras ψ : ∧ V1 ⊗ ∧V2 ⊗ ∧sV → ∧V ⊗ ∧V ⊗ ∧Vˆ induces a G-structure on (∧V1 ⊗ ∧V2 ⊗ ∧sV, D). As in the nonequivariant case, a relative model for (p0 , p1 ) is given by the inclusion (∧V1 ⊗ ∧V2 , D) → (∧V1 ⊗ ∧V2 ⊗ ∧sV, D) . I is then given by the tensor product A Sullivan model for MA

(∧V ⊗ ∧sV, δA ) = (∧V, d) ⊗(∧V1 ⊗∧V2 ,D) (∧V1 ⊗ ∧V2 ⊗ ∧sV, D) , where the structure of (∧V1 ⊗ ∧V2 )-module on (∧V, d) comes from the morphism (1, A) : (∧V1 , d) ⊗ (∧V2 , d) → (∧V, d).

221

222

5 : Geodesics

We extend the coefficient field to C and we take a basis of J formed by eigenvectors x with eigenvalue λx = 1. In this case δA (sx) − (λx x − x) is a decomposable element, which shows that the ideal I is acyclic. The commutative differential graded algebra (∧V ⊗ ∧sV, δA ) is therefore quasiisomorphic to the quotient A A A ¯ ¯ ⊗ A (∧V A ⊗ ∧s(V A ), δ¯A ) = (∧V A , d) (∧V ⊗∧V A ) (∧V1 ⊗ ∧V2 ⊗ ∧s(V ), D) . 1

2

This cdga is the minimal model of the free loop space on a space whose ¯ This proves the lemma. minimal model is (∧V A , d). Corollary 5.28 Let M be a compact simply connected Riemannian manifold and A an isometry. If for some q, (πq (M) ⊗ Q)A = 0, then M admits a nontrivial A-invariant geodesic. I has model (∧V A ⊗ ∧s(V A ), δ) ¯ by Lemma 5.27. Since Proof The space MA I A q A (πq (M) ⊗ Q) is dual to (V ) , we have πq (MA ) = 0 and the result follows from Theorem 5.23.

5.5

Existence of infinitely many A-invariant geodesics

A result of Tanaka extends the Gromoll–Meyer Theorem and defines a bridge between the existence of infinitely many A-invariant closed geodesics I. and the homology of MA Theorem 5.29 ([247]) Let M be a compact simply connected RiemanI are unbounded, then M admits nian manifold. If the Betti numbers of MA infinitely many geometrically distinct A-invariant geodesics. We now can use the minimal model described in Lemma 5.27 to give a I to be unbounded. criterion for the Betti numbers of MA Theorem 5.30 ([123]) Let M be a simply connected compact Riemannian manifold, let A be an isometry, and let (π∗ (M)⊗ Q)A denote the A-invariant part of rational homotopy. If dim (πodd (M) ⊗ Q)A ≥ 2, then the Betti I are unbounded, and M admits infinitely many nontrivial numbers of MA geometrically distinct A-invariant geodesics. I in Proof It follows from the description of the minimal model of MA I Lemma 5.27 that MA has the rational homotopy type of the free loop space on a space whose rational homotopy is (π∗ (M)⊗ Q)A . We then see from the I are Sullivan–Vigué theorem (Proposition 5.14) that the Betti numbers of MA unbounded. We then apply Tanaka’s Theorem 5.29 to infer the existence of infinitely many geometrically distinct A-invariant geodesics.

5.6

Gromov’s estimate and the growth of closed geodesics

Theorem 5.31 ([122]) Let M be a simply connected compact Riemannian manifold, and let A be an isometry. If M is rationally hyperbolic, then I are unbounded, and M admits infinitely many the Betti numbers of MA geometrically distinct A-invariant geodesics. Proof Let (∧V, d) be the model of a space of finite rational category (Definition 2.34) and suppose A is an automorphism of (∧V, d) of finite order ¯ also has k. Then we will prove that V A is infinite dimensional and (∧V A , d) finite category. ¯ is a quotient of (∧V, d), the second assertion follows Since (∧V A , d) from Theorem 2.81. By Theorem 2.82 this will imply that (V A )odd is also infinite dimensional. This means that πodd (M) ⊗ Q contains infinitely many invariant elements. We then apply Theorem 5.30 to infer the existence of infinitely many geometrically distinct A-invariant geodesics. We extend the coefficient field to C. We suppose Ak = 1 and we first give the proof when k is a prime number p. Then we will only have to k

write k = k p where p is a prime number. We deduce that V A is infinite dimensional and by induction that V A is infinite dimensional. Therefore, consider the case k = p. Since the category of (∧V, d) is finite, by Theorem 2.82, there is some integer N such that for each α ∈ π∗ (M)⊗Q in degree > N, there is β with adm (α)(β) = [α, [α, . . . , [α, β] . . .]] = 0 for all m . We can suppose that α and β are eigenvectors of A having respective eigen2πi

values λq and λr with λ = e p . When mq + r ≡ 0 (mod p), then adm (α)(β) is an eigenvector with eigenvalue 1. Since we can do this for all α and M is hyperbolic, this proves that V A is infinite dimensional.

5.6


Once we know that manifolds have many closed geodesics, we can start to study the exact nature of the geodesics. A first thing to ask is whether the growth rate of the number of geodesics is understandable in terms of length, say. With this in mind, let M be a Riemannian manifold and denote by nT the number of geometrically distinct closed geodesics of length less than or equal to T. The behavior of the sequence nT is once again related to the Betti numbers of the free loop space, bi (LM). An important result on nT was obtained by Gromov for bumpy metrics, so let’s review this notion now.

223

224

5 : Geodesics

First, recall from Morse theory that, for a smooth function f : W → R on a manifold W, a submanifold N is called a nondegenerate critical submanifold if N is composed of critical points and if the Hessian is nondegenerate at each point in the direction normal to N. We are interested in certain submanifolds of LM; namely, recall that, for a closed geodesic c : S1 → M and an element of the circle t ∈ S1 , we obtain a new (but not geometrically distinct) closed geodesic c˜ defined by c˜ (s) = c(s + t). The circle S1 therefore acts on any such c to give a submanifold of LM, S1 (c). Definition 5.32 A metric is called bumpy if for each critical point c of the energy function, the manifold S1 (c) is a nondegenerate critical submanifold of LM. In fact, this is a generic condition. Indeed, as shown in [12], for each manifold, the set of bumpy metrics is dense in the Frechet space of all the possible metrics on M. The theorem of Gromov, improved by Ballmann and Ziller, can now be stated. Theorem 5.33 ([18], [119]) Let M be a simply connected compact manifold. Then for any bumpy metric and any field lk, there exists a > 0 and c > 0 such that nT ≥ a · max dim H p (LM; lk) . p≤cT

Proof sketch The idea of Gromov’s proof is the following. Denote by (LM)T the subspace of LM consisting of curves of length less than or equal to T. Let d(T) be the maximal integer d such that, for i = 0 . . . , d, the inclusion (LM)T → LM induces a surjective map in lk-homology in degree i. Gromov then proved that there are constants c and C depending only on the manifold such that cT ≤ d(T) ≤ CT. Now, when the metric is bumpy, Morse theory shows that (LM)T has the homotopy type of a CW complex with the number of cells in degree p equal to the number of closed geodesics of length less than or equal to T and with (Morse) index p. Therefore the number of closed geodesics of length ≤ T, nT , satisfies the following inequalities. nT ≥

d(T) i=1

bi (YT ) ≥

d(T) i=1

bi (LM) ≥

cT

bi (LM).

i=1

Moreover, Gromov conjectured that, for almost all manifolds M, the sequence nT has exponential growth. One important result in this direction is due to P. Lambrechts.

5.6


Theorem 5.34 ([164]) Let M and N be simply connected compact manifolds of the same dimension n and let P be their connected sum, P = M#N. Suppose that, for some field lk, H ∗ (M; lk) is not singly generated as an algebra and H ∗ (N; lk) is not the cohomology of a sphere. Then the sequence nT (for P) has exponential growth. Remark 5.35 When π1 (M) and π1 (N) are not finite, then by Example 5.6, M#N admits infinitely many geometrically distinct closed geodesics. Theorem 5.34 is a generalization of that result to the case when the manifolds are simply connected. Although the result of Lambrechts is true for any field, we will restrict ourselves to a proof in the rational setting. First, let’s briefly recall the construction of the model of the connected sum AM#N (also see Example 3.6). Let (∧V, d) and (∧W, d) be the minimal models for manifolds M and N of the same dimension n. Denote by S a complement of the cocycles in degree n in ∧V and by I the acyclic differential ideal I = S ⊕ (∧V)>n . We denote by AM = (∧V)/I the quotient of ∧V by the acyclic ideal. Since I is acyclic, the quotient homomorphism is a quasi-isomorphism (∧V, d) → AM . We do the same for N to obtain a (nonminimal) model AN for N. Denote by uM and uN cocycles in degree n representing the fundamental cohomology classes in AM and AN . We deduce from Example 3.6 that a model for the connected sum M#N is given by AM#N = ((AM ⊕Q AN ) ⊕ Q · z, d)

with

d(z) = uM − uN .

We will compute the cohomology of the free loop space on M#N by using Hochschild homology. Hochschild homology is defined in Section 5.9, where we recall the fundamental isomorphism H ∗ (L(M#N); Q) ∼ = HH∗ (APL (M#N)) . Since Hochschild homology transforms quasi-isomorphisms of differential graded algebras into isomorphisms, we deduce the isomorphism H ∗ (L(M#N); Q) ∼ = HH∗ (AM#N ) . The proof of Lambrechts’s theorem is now a direct corollary of the following proposition. Proposition 5.36 The sequence dim HHq (AM#N ) in Hochschild homology has exponential growth. Proof Let x and x be two indecomposable elements in A 0, we see chaos in the orbit structure of φ. Of course, the definition of topological entropy applies to any flow, but it is the geodesic flow which seems to hold within it homotopical

227

228

5 : Geodesics

information. A result of A. Manning (see the discussion in [224, Chapter 5] and Section 6.1 for the definition of sectional curvature secM ) says that, for −κ ≤ secM ≤ κ with respect to a metric g, √ htop (g) ≤ (dim (M) − 1) κ. In particular, flat manifolds have htop (g) = 0. It is perhaps not surprising that the entropy of the geodesic flow is related to the growth in the number of geodesics joining two points when length is increased. Indeed, a result of Mañé characterizes htop (g) in this way, showing that entropy is truly a manifestation of the underlying Riemannian structure alone. With this in mind, let nT (x, y) denote the number of geodesic segments from x ∈ M to y ∈ M of length ≤ T with respect to the metric g. More precisely, for fixed x ∈ M, let DT M = {(x, v) ∈ Tx M | |v| ≤ T} and define nT (x, y) = #{exp−1 (y) ∩ DT M}, where expx : Tx M → M is the exponential map. (Note that the definition of nT (x, y) just says that you count the number of vectors v of length ≤ T such that expx (v) = y. Two vectors can be collinear and counted as distinct if they have different lengths.1 ) Then we have (see [224, Corollary 3.22], for instance) Theorem 5.39 The topological entropy htop (g) obeys the following inequality: for fixed x ∈ M, 1 log nT (x, y) dy . htop (g) ≥ lim sup M T→∞ T This result is proved by using a theorem of Yomdin which shows that the relative growth rate of volume under the geodesic flow is majorized by topological entropy. The inequality in Theorem 5.39 allows us to relate the algebraic topology of M to topological entropy via the following result of M. Gromov (generalizing his result for closed geodesics). Theorem 5.40 There exists a constant C > 0 such that, for fixed x, y ∈ M, nT (x, y) ≥

[C(T−1)]

bi (M),

i=1

where bi denotes the ith Betti number and [−] denotes the greatest integer function. 1 Thanks to Gabriel Paternain for making this point clear to us.

5.7

The topological entropy

The ideas behind this result are very similar to what we did leading up to Theorem 5.7. The energy functional E may be defined on the space of paths in M from x to y, M (x, y) by 1 1 c (t), c (t) dt . for c ∈ M (x, y), E(c(t)) = 2 0 and it can be shown that the critical points of E are geodesics joining x and y. Furthermore, these critical points are nondegenerate, so Morse theory can be applied. In fact, as usual, Morse theory is applied to “finite pieces” − −1 2 of loop spaces by defining, for example, T M (x, y) = E (−∞, T /2) and proving that critical points of E correspond to geodesic segments of length less than T. The Morse inequalities say that the sum of the Betti numbers to a given degree are majorized by the number of critical points up to that index. Furthermore, Morse theory tells us that, for a closed Riemannian manifold M and nonconjugate points x, y ∈ M (see Exercise 5.4), M (x, y) has the homotopy type of a countable CW-complex with one cell of dimension k for each geodesic from x to y of index k. Recall that y is conjugate to x along a geodesic α with α(0) = x if the exponential map expx : Tx M → M has expx (tA) = α(t) and y is a critical value for expx (tA). Of course, α is a critical point for the energy E, so we can also consider the Hessian of E (in a suitable sense) at α. The dimension of the nullspace of this Hessian is called the multiplicity of the conjugate point y. Now, the index of a geodesic from x is the index of the Hessian of the energy (i.e. the maximum dimension of a subspace on which the Hessian is negative definite) and this is the number of conjugate points to x along the geodesic (counted with multiplicities). All of these concepts may be found in [195]. The final thing to notice is that M (x, y) has the homotopy type of the ordinary loop space M. To see this, observe that both M and M (x, y) are fibers of the “smooth” path space (Serre) fibration p : P(M) → M, where P(M) = {α : I → M α(0) = x, and α is smooth},

p(α) = α(1).

Namely, M = p−1 (x) and M (x, y) = p−1 (y). Therefore, since fibers of a fibration have the same homotopy type, we have M M (x, y). Alternatively, if γ is a path from x to y, then composition with γ and with the inverse path γ¯ gives inverse homotopy equivalences between M and M (x, y). In particular, the Betti numbers of M (x, y) are the same as those of M. These identifications then give the inequality of Theorem 5.40. Note the following immediate application (due to Serre; see [236]).

229

230

5 : Geodesics

Theorem 5.41 If Mn is a closed simply connected Riemannian manifold and x, y ∈ M are nonconjugate points, then lim nT (x, y) = ∞.

T→∞

Proof Suppose limT→∞ nT (x, y) is finite. Then, by Theorem 5.40, M (x, y) has the homotopy type of a finite complex. Now, since H n (M; Q) = 0, the minimal model (∧V, d) is nontrivial and contains a generator of odd degree because, otherwise, the dimension of H ∗ (M; Q) is infinite. Therefore, πk−1 (M)⊗ Q = πk (M)⊗ Q = 0 for some odd integer k. Then the minimal model of M has an even degree generator u. Since the differential of the model of M is zero, we see that us is a nontrivial cocycle for all s ≥ 0, and this contradicts the finiteness of M (x, y) M. Now, let’s return to entropy and combine the estimates of Theorems 5.39 and 5.40 to see how properties of the geodesic flow on a manifold serve to constrain its homotopy type. Theorem 5.42 ([223, Theorem 3.2]) If (M, g) is a simply connected closed Riemannian manifold with htop (g) = 0, then M is rationally elliptic. [C(T−1)] Proof By Theorem 5.40, we have nT (x, y) ≥ bi (M). If we i=1 1 integrate over M, take lim sup T log and apply Theorem 5.39, we obtain 1 htop (g) ≥ lim sup log nT (x, y) dy M T→∞ T ⎛ ⎞ [C(T−1)] 1 ≥ lim sup log ⎝vol(M) bi (M)⎠ . T→∞ T i=1

By hypothesis, htop (g) = 0, so we must have 2m 3 1 bi (M) = 0. lim sup log m→∞ m i=1

But this means that the sum of the Betti numbers of the loop space grows sub-exponentially. By Theorem 2.74 and Theorem 2.75, this is equivalent to M being rationally elliptic. Example 5.43 In [223], Paternain provided an important example of a situation where the geodesic flow has htop (g) = 0. Let (Mn , g) be a closed manifold with metric g. Although the cotangent bundle T ∗ (M) has a canonical symplectic structure, the tangent bundle obtains one only through the duality provided by the metric. The geodesic flow is then the Hamiltonian

5.7

The topological entropy

flow (with respect to the symplectic structure associated to g) corresponding to the Hamiltonian provided by the g-norm, H = − : T(M) → R. The geodesic flow is completely integrable as a Hamiltonian system if there are n smooth functions on T(M), f1 = H, f2 , f3 , . . . , fn such that {fi , fj } = 0, where {−, −} is the Poisson bracket on functions given by the symplectic structure. Completely integrable Hamiltonian systems have a very particular structure which allows the flow to be analyzed by restricting to invariant tori in T(M). To each fj , we may associate a vector field Xj by using the nondegeneracy of the symplectic form ω: iXj ω = dfj , where iXj is interior multiplication by Xj . Of course, the vector field Xj has an associated flow φjt . If, for each j = 1, . . . , n, the flow φjt actually is the orbit flow given by a circle action, then we say that the geodesic flow is completely integrable with periodic integrals. Essentially, we are saying that the Hamiltonian system has T n as a symmetry group, thus reducing the system to solution by integration (i.e. quadrature). We then have the following. Theorem 5.44 ([223]) Let (M, g) be a simply connected closed Riemannian manifold whose geodesic flow is completely integrable with periodic integrals. Then htop (g) = 0 and M is rationally elliptic. Now let M be a compact simply connected Riemannian manifold and let N be a compact simply connected submanifold. Fix p ∈ M and T > 0. Paternain and Paternain (see [225]) define nT (p, N) to be the number of geodesics leaving orthogonally from N and ending at p with length ≤ T. If p is not a focal point of N, nT (p, N) is finite. We can therefore define nT (p, N)dµ , IN (T) = M

where µ is the measure induced by the Riemannian structure. Then extending the method of Gromov, Paternain and Paternain prove Theorem 5.45 ([225]) There exists a constant C > 0 depending only on the geometry of M and N such that if (p, N) denotes the Hilbert manifold of paths from N to p, then for k ≥ 1, k i=1

bi ((p, N)) ≤

1 IN (Ck) . Vol M

Corollary 5.46 Suppose that M is rationally hyperbolic and N is rationally elliptic, then the sequence IN (T) has exponential growth.

231

232

5 : Geodesics

Proof We have a fibration M → (N, p) → N × {p} from which we deduce the homotopy fibration N → M → (N, p) . Since the homology of M has exponential growth and the homology of N polynomial growth, the homology of (N, p) must have exponential growth.

5.8

Manifolds whose geodesics are closed

The topological entropy of a smooth diffeomorphism is essentially a measure of the volume growth induced by its iterates. In [100], various types of volume growth invariants are defined and related to the rational ellipticity or hyperbolicity of the manifold of definition. Here, in the spirit of this chapter, we want to discuss a fact that is implicit in [27], but seemingly not observed until its appearance in [100], concerning manifolds with the property that all geodesics are periodic. The basic reference for these manifolds is [27]. Definition 5.47 Let Mn be closed Riemannian manifold and x ∈ M. 1. A geodesic (always parametrized by arclength) issuing from x, α : [0, T] → M is periodic with period T if α(0) = x = α(T) and α(0) ˙ = α(T). ˙ 2. M is a PTx -manifold if all geodesics issuing from x are periodic with common period T. (Note that some geodesics are allowed to come back to x before length T.) 3. If M is a PTx manifold for all x ∈ M, then M is said to be a P-manifold. (It is a fact that if all the geodesics of a manifold are periodic, then they admit a common period. See [27, Lemma 7.11].) 4. The index of the geodesic α : [0, T] → M is the number of t ∈ (0, T) such that α(t) is conjugate to α(0) = x along α. (Note that α(T) is not counted here. This is explained by Proposition 5.48.) Proposition 5.48 ([27, 7.25]) If Mn is a PTx -manifold, then x is conjugate to itself with multiplicity n − 1 along any geodesic from x to x of any length. Furthermore, along any geodesic from x to x of length T, there are k conjugate points for some fixed integer k. Now here are the facts about PTx -manifolds culled from [27, Chapter 7]. First, note that there always exists some point y ∈ M that is not conjugate to x along any geodesic. This is a simple consequence of Sard’s theorem since conjugate points are critical values of the exponential map expx (see

5.8

Manifolds whose geodesics are closed

Exercise 5.4). Then, it can be shown that there are only a finite number of geodesics, α1 , . . . , αN from x to y of length strictly less than T. To see this, suppose αi , i = 1, . . . , ∞ is an infinite set of geodesics from x to y with lengths ti smaller than T and restrict to the compact set S(Tx M) × [0, T] (where S(Tx M) denotes the unit sphere in Tx M) so that there is a convergent subsequence of the sequence (αi (0), ti ) which converges to the limit (α (0), t). Then lim αi (ti ) = α(t) = y is conjugate to x along the geodesic α with initial tangent vector α (0) (see Exercise 5.4). This is a contradiction, so only a finite number of geodesics exist. So now, let k1 , . . . , kN be the indices of the geodesics α1 , . . . , αN from x to y of length strictly less than T. Without loss of generality, let k1 ≤ k2 ≤ · · · ≤ kN . There is some minimizing geodesic of length d(x, y), so the energy Hessian is positive semidefinite along it and, thus, k0 = 0. Now, since M is a PTx -manifold and the lengths of the αi are all less than T, we know there is an opposite geodesic αi : [0, T − d(x, y)] → M given by αi (t) = αi (T − t). (Think of the sphere. A minimal geodesic is the shorter part of a great circle, while the longer part of the geodesic is the opposite geodesic.) Then, since the indices are the numbers of conjugate points and all loops have index k, we see that the opposite geodesic is one of the αj and that its index is k − ki . In particular, the opposite geodesic to the minimizer then has largest index kN = k (by Proposition 5.48). The crucial fact is that every geodesic from x to y is the union of p geodesics from x to x of length T (for arbitrary p) and one of the αi ’s. Of course, each geodesic from x to y has k + n − 1 conjugate points counted with multiplicity. Therefore, for each integer p, there are exactly N geodesics from x to y whose lengths are between p T and (p + 1)T and whose indices are ki + p (k + n − 1). Note that, since k is the largest index, different p produce a completely different set of indices. If we now consider M (x, y), we see that, for each integer m, the number of critical points (i.e. geodesics) with fixed index m is bounded by N. Since each critical point gives a cell in a CW-decomposition of M (x, y), the number of cells, and hence the Betti number, in each dimension is also bounded by N. Because M (x, y) M, we have the following. Theorem 5.49 If M is a PTx -manifold, then dim H q (M; Q) ≤ N for some fixed integer N and all q. This holds in particular for P-manifolds. Now, an argument similar to that in the proof of Proposition 5.14 shows that the Betti numbers of the loop space of a simply connected finite complex are bounded if and only if the rational cohomology ring of the complex is singly generated (see Exercise 5.5). Therefore, in particular, we have

233

234

5 : Geodesics

Corollary 5.50 If M is a PTx -manifold, then H ∗ (M; Q) is singly generated. Hence, M is rationally elliptic. This holds in particular for P-manifolds. Therefore, we see how the nature of the geodesics on a manifold can have powerful consequences for its homotopy structure.

5.9

Bar construction, Hochschild homology and cohomology

The bar construction and Hochschild homology play important roles in geometric applications of algebraic models. One reason for this is that Hochschild homology is inextricably linked to the cohomology of the free loop space. Furthermore, it will be an important tool for the proof of Lambrechts’s result on the cohomology of the free loop space of a connected sum of varieties, as well as for the definition of iterated integrals (see Section 9.6) and for the computation of the Chas–Sullivan loop product (see Section 8.4). Indeed, in Subsection 8.4.5, we shall see some implications that the Chas–Sullivan loop product has for closed geodesics. Let (A, d) be a differential graded algebra over a field lk, A = ⊕∞ Ak , k=0 ¯ = Ker ε. We with an augmentation ε : A → lk, and augmentation ideal A q q+1 ¯ ¯ ¯ ¯ denote by sA the suspension of A, (sA) = (A) . Definition 5.51 The double bar construction on (A, d) is the complex (B(A, A, A), D) = (⊕n Bn (A, A, A), D) ¯ ⊗ · · · ⊗ sA) ¯ ⊗ A, (n copies of sA). ¯ The element with Bn (A, A, A) = A ⊗ (sA a ⊗ (sa1 ⊗ · · · ⊗ san ) ⊗ a is denoted by a[a1 | · · · |an ]a . The differential D decomposes into two terms D = b + d. b(a[a1 | . . . |an ]a ) = −âa1 [a2 | . . . |an ]a −

n−1

(−1)i aˆ [â1 | . . . |âi ai+1 | . . . |an ]a

i=1

+ (−1)n−1 aˆ [â1 | . . . |ân−1 ]an a , d(a[a1 | . . . |an ]a ) =d(a)[a1 | . . . |an ]a +

n (−1)i aˆ [â1 | . . . |âi−1 |d(ai )|ai+1 | . . . |an ]a i=1

+ (−1)n aˆ [â1 | . . . |ân ]d(a ) . Here aˆ = (−1)|a| a.

5.9


The complex B(A, A, A) is a right and a left differential A-module, quasiisomorphic to A as an A-bimodule. Therefore for each right A-module M, and each left A-module N, we can form the complex B(M, A, N) = M ⊗A B(A, A, A) ⊗A N . ¯ (A) = B(Q, A, Q) is a coalgebra. The When M = N = Q, then B comultiplication is defined by ¯ → T(sA) ¯ ⊗ T(sA) ¯ , ∇ : T(sA) ∇[a1 |a2 | . . . |an ] =

n [a1 | . . . |ai ] ⊗ [ai+1 | . . . |an ] . i=0

We have ¯ (A)) = H ∗ (B(Q, A, Q)) is Theorem 5.52 ([1]) If A = C ∗ (X; Q), then H∗ (B isomorphic to H ∗ (X; Q) as a coalgebra. When M = A and N = Q the complex B(A, A, Q) is contractible and is therefore a good model for the path space PX. Our last example is related to the free loop space. The multiplication A ⊗ A → A also makes (A, d) into a left and a right A-module. Definition 5.53 The Hochschild complex B∗ (A) is the tensor product ¯ = (A, d) ⊗A⊗A B(A, A, A) B∗ (A) := A ⊗ T(sA) where T(V) denotes the tensor algebra on V. Its homology is denoted by ¯ is the sum of an internal HH∗ (A). The induced differential D on A ⊗ T(sA) differential d and an external one b defined by d(a[a1 | . . . |an ]) = d(a)[a1 | . . . |an ] +

n

(−1)i aˆ [â1 | . . . |âi−1 |d(ai )|ai+1 | . . . |an ] ,

i=1

b(a[a1 | . . . |an ]) = −âa1 [a2 | . . . |an ] −

n−1

(−1)i aˆ [â1 | . . . |âi ai+1 | . . . |an ]

i=1

− (−1)n+|an |(|a|+|a1 |+...+|an−1 |+n−1) an aˆ [â1 | . . . |ân−1 ] . The relation between the homology of the free loop space and the Hochschild homology has been proved by different authors and can be found in [47], [109] and [149].

235

236

5 : Geodesics

Theorem 5.54 If X is a finite simply connected space with finite Betti numbers, then H ∗ (LX; lk) ∼ = HH∗ (C ∗ (X; lk)) , where C ∗ (X; lk) denotes the usual singular cochain algebra of X with coefficients in the field lk. It is a classical result that the Hochschild homology is invariant along quasi-isomorphisms of differential graded algebras (see [109]), so we have Corollary 5.55 If X is a simply connected space with minimal model (∧V, d), then H ∗ (LX; Q) ∼ = HH∗ (APL (X)) ∼ = HH∗ (∧V, d) . When M is a manifold, we then have Corollary 5.56 If M is a simply connected compact manifold with de Rham algebra of forms ADR (M) and rational minimal model (∧V, d), then H ∗ (LM; R) ∼ = HH∗ (ADR (M)) ,

H ∗ (LM; Q) ∼ = HH∗ (∧V, d) .

In the rational case, the isomorphism between the Hochschild homology of the minimal model (∧V, d) of M and the cohomology of the free loop space is quite easy to understand. Indeed, the bar construction B((∧V, d), (∧V, d), (∧V, d)) and the differential algebra (∧V ⊗ ∧V ⊗ ∧sV, D) constructed in the proof of Theorem 5.11 are quasi-isomorphic semifree (∧V ⊗ ∧V)-modules. If we take the tensor product with (∧V, d) over (∧V ⊗ ∧V), we also obtain quasi-isomorphic complexes (see Exercise 2.7). The first tensor product gives the Hochschild homology and the second one is the model of the free loop space (Theorem 5.11). As we saw, when A = C ∗ (M; lk), the double bar construction leads to complexes that give the homology of the various usual path spaces. Table 5.1 is a summary of the connections. In Section 8.4 we will also need the Hochschild cohomology algebra of a differential graded algebra. We recall the definition here. Table 5.1 Bar constructions and spaces.

Bar constructions

Spaces

B(A, A, A)

M[0,1]

B(A, A, lk)

P(M)

¯ (A) = B(lk, A, lk) B

(M)

B(A)

LM

5.9


Definition 5.57 The Hochschild cohomology of a graded differential algebra (A, d), HH ∗ (A, A), is the cohomology of the complex HomA (B(A), A). This cohomology is a graded algebra. The product " of two elements ¯ A) = HomA (B(A), A) is defined by f , g ∈ Hom(T(sA), f ⊗g ∇ m ¯ ⊗ T(sA) ¯ −→ ¯ −→ T(sA) A ⊗ A −→ A . f " g : T(sA)

Exercises for Chapter 5 Exercise 5.1 Construct the model for the free loop space on the manifolds S3 × S3 and CP(n) and deduce the cohomology algebra of the corresponding free loop spaces. Exercise 5.2 Let M be the usual torus and A be an isometry inducing in homotopy the morphism 0 −1 1 0 Show that M admits at least one nontrivial A-invariant geodesic. Exercise 5.3 Let G/K be a simply connected homogeneous space. Suppose that rank G− rank K is odd. Show that for every isometry A, there exists a nontrivial A-invariant geodesic. Exercise 5.4 For a smooth manifold M, the exponential map at a point p ∈ M, expp : Tp M → M is given by expp (tv) = γ (t), where γ : [0, L] → M is the unique unit-speed geodesic with γ (0) = p and γ (0) = v. A point q = γ (s) is conjugate to p if dexpp : Tsv Tp M = Tp M → Tq M is singular (i.e. has a kernel) at sv = sγ (0). The significance of conjugate points is mainly due to the fact that geodesics cannot minimize arclength beyond the first conjugate point. Classically, a conjugate point was said to be the intersection of infinitesimally close geodesics. The point of this exercise is to explain what this means in the context of the discussion following Proposition 5.48. Suppose αi : [0, ti ] → M is a sequence of unit-speed geodesics with length less than T, αi (0) = p and αi (ti ) = q. Because S(Tp M)×[0, T] is compact (where S(Tp M) is the unit sphere in Tp M), the sequence (αi (0), ti ) has a convergent subsequence converging to (α (0), t). This initial data gives a geodesic α with α(0) = p and α(t) = q. The claim now is that q must be conjugate to p along α. Show this using (and verifying) the following hints. (1) Suppose that q is not a conjugate point. Therefore, there is an open set U ⊂ Tp M with tα (0) ∈ U such that expp |U is a diffeomorphism onto an open neighborhood of q. (2) If K is large enough, then, for all i > K, ti αi (0) ∈ U. This is what is meant by infinitesimally close.

237

238

5 : Geodesics

(3) Show from the definition of the exponential map above that expp (ti αi (0)) = expp (tα (0)). (4) Why is this a contradiction? Exercise 5.5 Here is a warm-up for Proposition 5.14. Show that the Betti numbers of the loop space of a simply connected finite complex X are bounded if and only if its rational cohomology ring is singly generated. Hint: suppose H ∗ (X; Q) is not singly generated and note that (as in Proposition 5.14) there are always at least two odd degree generators in the minimal model for X. Then, since the differential in the model of the loop space is zero, there are two even generators u and v that are cocycles. If the respective degrees are 2k and 2s with L = lcm(2k, 2s), L = 2kp, L = 2sq, then show that dim H mL (X; Q) ≥ m + 1 since up i vq j ∈ H mL (X; Q) for i, j ≥ 0 with i + j = m.

6

Curvature

Recently algebraic models have found a place in differential geometry, especially in questions revolving around types of curvature restrictions. In particular, F. Fang and X. Rong [84] and B. Totaro [256] have approached certain questions of Karsten Grove in this spirit and the work of Belegradek and Kapovitch [23] on the existence of non-negative curvature metrics on vector bundles has also relied on models. Of course, when we speak of algebraic models here, we are speaking of models ranging from simple cohomology all the way to minimal models and relative models. In this chapter, we want to examine these applications of models to curvature questions as well as other relevant connections between Riemannian geometry and algebraic models. A good general reference for Riemannian geometry and, especially, various finiteness theorems for diffeomorphism or homotopy type and topics surrounding Gromov–Hausdorff convergence, is [226].

6.1

Introduction: Recollections on curvature

A smooth n-manifold Mn is Riemannian if each tangent space Tp (M) has a Euclidean metric −, − and these vary smoothly on M. Associated to the Riemannian metric is a covariant derivative ∇ called the Riemannian connection. Proposition 6.1 Let V, Z and W be tangent vector fields on M and f : M → R be a function on M. The following are properties of the covariant derivative of M: (i)

∇V (Z + W) = ∇V Z + ∇V W;

(ii)

∇fV Z = f ∇V Z;

(iii)

∇V fZ = V[f ] Z + f ∇V Z;

240

6 : Curvature

(iv)

∇V Z − ∇Z V = [V, Z], where [−, −] is the usual bracket of vector fields;

(v)

V(Z, W) = ∇V Z, W + Z, ∇V W, where the vector field V acts on the function Z, W in the standard “directional derivative” way and −, − denotes the metric on M.

In fact, property (v) may be used to define the covariant derivative and we can also then see that the Riemannian connection is unique. We can use the connection to make the following. Definition 6.2 Let X, Y and Z be vector fields on the manifold M. The Riemann curvature is defined to be R(X, Y)Z = ∇[X,Y] Z + ∇Y ∇X Z − ∇X ∇Y Z. A more intuitive version of curvature that still holds within it all of the information provided by R is the sectional curvature, sec(X, Y) =

R(X, Y, X, Y) X, XY, Y − X, Y2

,

where we use the notation def

R(X, Y, Z, W) = R(X, Y)Z, W. The sectional curvature may be interpreted as follows. Suppose X and Y span a tangent plane inside the tangent space Tp (M). The exponential map carries the plane to a surface inside the manifold, so the ordinary Gauss curvature (see, for instance, [216, page 115, Chapter 3]) of the surface may be computed. This Gauss curvature is then the sectional curvature sec(X, Y). A more technical approach to sectional curvature is to consider it as a function on the Grassmannian bundle associated to the tangent bundle. More specifically, at x ∈ Mn , sectional curvature is a function sec : G2 (Tx (M)) → R, where G2 (Tx (M)) denotes the Grassmann manifold of 2-planes in Tx (M). We can construct a bundle with fiber G2,n = G2 (Rn ) by noting that G2,n = O(n)/(O(2) × O(n − 2)) and then using the natural O(n) action on G2,n to form an associated bundle, G2,n (M) = F(M) ×O(n) G2,n to the principal (frame) bundle O(n) → F(M) → M underlying the tangent bundle of M. Then sectional curvature becomes a function sec : G2,n (M) → R. Furthermore (see, for instance, [28, pages 166–167]), the function sec is smooth and, therefore, continuous. If M is a closed manifold, then G2,n (M) is compact, so its image under sec is also compact and we obtain the following.

6.1

Introduction: Recollections on curvature

Proposition 6.3 If M is a compact manifold, then there exist λ, κ ∈ R such that for any plane P ⊂ Tx (M) and for any x ∈ M, κ ≤ secP ≤ λ. Now, if we scale the metric by multiplying by a constant (i.e. −, − → c−, −, c > 0), then sectional curvature is divided by that constant. In this way, for compact manifolds, we can always consider Im(sec) ⊂ (−1, 1) if we desire. The price we pay for this scaling is that the diameter of the manifold increases. So, there must always be a sort of tension between sectional curvature and diameter. In fact, we shall see this reflected in the hypotheses of the main results below. Once an invariant such as sec is defined, it is standard procedure to restrict the invariant in some way and try to classify the manifolds obeying the restriction. This classification is often a topological one. For example, Hadamard’s famous theorem implies that a manifold M with sec(X, Y) ≤ 0 for all pairs X, Y must be an Eilenberg–Mac Lane space, M = K(π , 1). Similarly, the famous pinching theorem says that bounding sectional curvature as 1/4 < sec ≤ 1 implies that the manifold is a sphere – a very strong constraint indeed! Example 6.4 A manifold Mn with constant sectional curvature equal to zero is said to be flat; it is also called a Bieberbach manifold since it was L. Bieberbach who characterized such a manifold as one covered by Euclidean space Rn such that the integral translations Zn are a normal subgroup of finite index in π1 (M). In other words, there is a short exact sequence of groups,

Zn → π1 (M) → F where F is finite. Such a group is called a Bieberbach group. It is a fact that flat manifolds are classified up to diffeomorphism by their fundamental groups, so in this case group theory determines topology. Example 6.5 If G is a compact Lie group (see Chapter 1), then G has a bi-invariant metric −, − so that geodesics with respect to this metric are the one-parameter subgroups arising from integration of left invariant vector fields. Therefore, for any Y ∈ g (where g is the Lie algebra of G), ∇Y Y = 0, where ∇ is the associated Riemannian connection. Now, by expanding ∇X−Y (X−Y) = 0, we obtain ∇X Y = −∇Y X. Then, since ∇ is the Riemannian connection, we have [X, Y] = ∇X Y − ∇Y X = 2∇X Y. Putting this in Definition 6.2, we easily calculate (with the help of the Jacobi identity) R(X, Y)Z = 14 [[X, Y], Z]. Substituting this result into the definition of sec gives sec(X, Y) =

1 |[X, Y]|2 , 4

241

242

6 : Curvature

where | − | is the norm associated to the metric and X, Y are orthonormal. See Exercise 6.2 for details. Thus, compact Lie groups have non-negative sectional curvature. Once we know this, we can create more manifolds with non-negative sectional curvature by using the following result. Proposition 6.6 (O’Neill’s formula: [213]) Let π : M → N be a Riemannian submersion and suppose that X and Y are orthonormal horizontal vector fields on M. Then sec(π∗ X, π∗ Y) = sec(X, Y) +

3 |[X, Y]v |2 4

where [−, −]v denotes the vertical projection. (If X and Y are not orthonordef

mal, then the second term on the right must be divided by |X ∧ Y| = X, X Y, Y − X, Y2 .)

Therefore, a Riemannian submersion can only increase sectional curvature, never decrease it. Using this, we can see that, for instance, homogeneous spaces G/H and biquotients G//H (where H is closed) also have non-negative sectional curvature metrics. Various other upper and lower bounds on sectional curvature may be combined with restrictions on, for instance, diameter and volume, to create new classes of manifolds. As an example, let M≤D ≥κ (n) denote the class of closed n-manifolds with sectional curvature bounded below by κ and diameter bounded above by D. Gromov proved that there are constants b(n, κ, D), depending on dimension, sectional curvature and diameter, such that the sum of the Betti numbers of any M ∈ M≤D ≥κ (n) is bounded above by b(n, κ, D). This puts a strong topological constraint on manifolds in M≤D ≥κ (n). If we also include a volume restriction, volume ≥ V with asso≤D,≥V ciated class M≥κ (n), then, for n ≥ 5, there are only finitely many diffeomorphism types depending on n, κ, D and V. Because of this result and various pinching theorems, the class M≤D κ≤sec≤λ (n) of closed n-manifolds with sectional curvature and diameter obeying κ ≤ sec ≤ λ and diam ≤ D has proven to be of some interest. In particular, K. Grove asked the following question. Question 6.7 (Grove’s question) Does the sub-class of M≤D κ≤sec≤λ (n) consisting of simply connected manifolds contain finitely many rational homotopy types? Let’s now consider this question and see how algebraic models play a key role.

6.2

6.2

Grove’s question

Grove’s question

In this section we consider the work of Fang and Rong [84] and of B. Totaro [256] on Grove’s conjecture. The Fang–Rong approach essentially creates the right algebra to contradict the conjecture and then realizes the algebra using minimal models. Totaro, on the other hand, uses biquotients to construct manifolds with the right curvature properties and then shows that a sufficient number of these also have the right algebra to contradict the conjecture. Since the methods are so different, we present both. Both approaches use a fundamental result due to J.-H. Eschenburg that relates qualities of horizontal projections associated to a certain sequence of submersions to the qualities of the resulting sectional curvatures. More specifically, suppose G, M and Mi (i = 1, 2, . . .) are Riemannian manifolds with submersions π : G → M and πi : G → Mi . For each g ∈ G, there are projections h(g) : Tg (G) → H(Tg (G)) ⊆ Tg (G) and hi (g) : Tg (G) → Hi (Tg (G)) ⊆ Tg (G) to the horizontal subspaces determined by the bundle structure of the respective submersions. We may then consider h and hi as elements of (End(T(G))), the space of sections of the endomorphism bundle associated to the tangent bundle of G. As such, we can consider the convergence properties of the sequence hi in the C 1 -topology on (End(T(G))). Then Eschenburg proved Theorem 6.8 [(81, Proposition 22]) If hi → h in the C 1 -topology on (End(T(G))), then for any g ∈ G and any horizontal 2-plane P ⊂ Tg (G), secMi (πi∗ P) → secM (π∗ P) and the convergence is uniform in P. Proof idea Let X and Y be π-horizontal vector fields which span P. Then hi (X) → X and hi (Y) → Y in the C 1 -topology. Then Proposition 6.6 and some simplification of the covariant derivative give secMi (πi∗ (X), πi∗ (Y)) = secG (hi (X), hi (Y)) +

3|(I − hi )(∇hi (X) hi (Y))|2 . |hi (X) ∧ hi (Y)|2

Then it is plain that secMi (πi∗ (X), πi∗ (Y)) converges to secM (X, Y) uniformly.

6.2.1

The Fang–Rong approach

In [84], Fang and Rong answered Grove’s question as follows. (Note that the formulation of the result provides an upper bound for diameter in order for the examples of the theorem to have sectional curvature scaled between −1 and 1.)

243

244

6 : Curvature

Theorem 6.9 For any n ≥ 22, there exists D > 0 such that the subclass of M≤D −1≤sec≤1 (n) consisting of simply connected manifolds contains infinitely many rational homotopy types. In order to prove this result, let’s first consider the algebra required. For any integer α > 0, let ∧α = (∧(x1 , x2 , y, zα ), d) be the freely generated commutative differential graded algebra with deg(x1 ) = deg(x2 ) = 2, deg(y) = 5, deg(zα ) = 7 and differential defined by d(x1 ) = d(x2 ) = 0,

d(y) = x21 x2 ,

d(zα ) = x41 + x42 + (αx1 + x2 )4 .

Lemma 6.10 If α = β ∈ Z, then H ∗ (∧α ) ∼ = H ∗ (∧β ). Proof The argument here is standard. Let’s suppose that there is an isomorphism f : H ∗ (∧α ) → H ∗ (∧β ), where α = β. We shall denote cohomology classes by their representatives: x1 instead of [x1 ] etc. Also, the generators of ∧β will be denoted by x˜ 1 , x˜ 2 , y˜ and z˜ β . Now, degree 2 cohomology is generated by the x’s, so we must have f (x1 ) = ax˜ 1 +bx˜ 2 and f (x2 ) = cx˜ 1 +ex˜ 2 , where a, b, c and e are rational numbers. Note that, because f is an isomorphism, coefficients a and b cannot both be zero and coefficients c and e cannot both be zero. Now, because of the y-differentials, in cohomology we have x21 x2 = 0 = 2 x˜ 1 x˜ 2 and these relations are the only ones in the respective degree 6 cohomologies. Thus, x˜ 31 , x˜ 32 and x˜ 1 x˜ 22 are linearly independent in H ∗ (∧β ). We then have f (x21 x2 ) = 0 (ax˜ 1 + bx˜ 2 )2 (cx˜ 1 + ex˜ 2 ) = 0 a2 cx˜ 31 + a2 ex˜ 21 x˜ 2 + 2abcx˜ 21 x˜ 2 + 2abex˜ 1 x˜ 22 + b2 cx˜ 1 x˜ 22 + b2 ex˜ 32 = 0 a2 cx˜ 31 + 2abex˜ 1 x˜ 22 + b2 cx˜ 1 x˜ 22 + b2 ex˜ 32 = 0, using x˜ 21 x˜ 2 = 0. By linear independence, we obtain a2 c = 0, 2abe + b2 c = 0 and b2 e = 0. We must consider various cases. If a = c = 0, then the third condition says that either b = 0 or e = 0, contradicting the fact that f is an isomorphism. If a = 0 and c = 0, then the second condition says that b = 0, again contradicting the fact that f is an isomorphism. Therefore, we must have a = 0. The first condition then implies that c = 0. Hence e = 0 since f is an isomorphism and so b = 0 by the third condition. We then have f (x1 ) = ax˜ 1 and f (x2 ) = ex˜ 2 .

6.2

Grove’s question

Now, the z-differentials provide relations in degree 8 cohomology as follows. x41 + x42 + (αx1 + x2 )4 = 0 (α 4 + 1)x41 + 4αx1 x32 + 2x42 = 0 using x21 x2 = 0. (Of course, the relation (β 4 + 1)x˜ 41 + 4β x˜ 1 x˜ 32 + 2x˜ 42 = 0 also holds in H ∗ (∧β ).) If we apply f to the relation, using f (x1 ) = ax˜ 1 and f (x2 ) = ex˜ 2 , we get (α 4 + 1)a4 x˜ 41 + 4αae3 x˜ 1 x˜ 32 + 2e4 x˜ 42 = 0. We want to use the fact that x˜ 41 and x˜ 1 x˜ 32 are linearly independent elements with the dependence of x˜ 42 on them given by the relation. So let’s simplify this dependence by writing the relation above as

a 4

a (α 4 + 1) x˜ 1 x˜ 32 + 2x˜ 42 = 0. x˜ 41 + 4α e e We then have two representations of 2x˜ 42 : 2x˜ 42 = −((α 4 + 1)

a 4 e

x˜ 41 + 4α

a e

x˜ 1 x˜ 32 )

2x˜ 42 = −((β 4 + 1)x˜ 41 + 4β x˜ 1 x˜ 32 ). By uniqueness of representations, we obtain

a 4

a = β. = (β 4 + 1) and α (α 4 + 1) e e (Note that, since α and β are both positive, we must have a/e > 0 as well.) If we replace β in the first equation by its expression in terms of α given in the second equation, we obtain

a 4 a 4

a 4 α4 + = α4 + 1. e e e Hence, we have (a/e)4 = 1. Since a and e are rational and a/e > 0, it must be the case that a = e. Putting this into the equation α(a/e) = β then gives α = β, contradicting our initial assumption that α = β. Hence, no isomorphism f can exist. So, if we can find simply connected manifolds in M≤D −1≤sec≤1 (n) with the cohomology algebras above, then Theorem 6.9 will be proved. It will come as no surprise then that the algebras ∧α that appear above arise as (part of) algebraic models for these manifolds. Let’s now see how this comes about.

245

246

6 : Curvature

Let M8 = (∧(x1 , x2 , x3 , y, z), d) with deg(xi ) = 2, i = 1, 2, 3, deg(y) = 5, deg(z) = 7 and differential defined by d(xi ) = 0,

d(y) = x21 x2 ,

d(z) = x41 + x42 + x43 .

We can explicitly find a Z-form for this algebra; that is, we can find a CW-complex whose homotopy groups are finitely generated (over Z) for which M8 is an algebraic model. The first step is to take CP(∞) × CP(∞) × CP(∞) = K(Z, 2)×K(Z, 2)×K(Z, 2) = K. This models the xi generators as integral cohomology classes. The second step is to take the principal fibration induced by the classifying map representing the integral cohomology class x21 x2 ∈ H 6 (K; Z) = [K, K(Z, 6)]. The pullback of the universal fibration K(Z, 5) = K(Z, 6) → PK(Z, 6) → K(Z, 6) via the classifying map gives a fibration K(Z, 5) → X → K and X is an integral model for the xi generators and the y generator of M8 . Finally, use the cohomology class x41 + x42 + x43 ∈ H 8 (X; Z) = [X, K(Z, 8)] to induce a principal fibration K(Z, 7) → E → X with the generator of π7 (K(Z, 7)) integrally representing z ∈ M8 . The space E is then a Z-form for M8 . For the following result, recall from Subsection 2.5.4 the notion of a Postnikov tower for a space and its relation to minimal models. Lemma 6.11 For each n ≥ 21, there is a closed simply connected n-manifold B with H 2 (B; Z) = Z3 whose minimal model MB has degree 8 Postnikov piece MB (8) = M8 . Proof Let E[10] denote the 10-skeleton of M8 ’s Z-form E. Because the 10-skeleton carries E’s homotopy type through dimension 9, we also have ME[10] (8) = M8 . But E[10] is a finite CW-complex of dimension 10, so we can embed E[10] in Rn+1 with n ≥ 21 and take a thickening or regular neighborhood N; that is, a compact (n + 1)-manifold with boundary of the same homotopy type as E[10]: N E[10]. Let B = ∂N be the closed n-manifold boundary of N and note that the 10-skeleton of B is E[10] by properties of thickenings (see [159, Section II]). Thus MB (8) = M8 . Now that we have the n-manifold B (n ≥ 21), we construct a closed (n + 1)-manifold Mα as a principal circle bundle over B classified by the Euler class eα = αx1 + x2 − x3 , where α ∈ Z+ and the xi are the integral generators of H 2 (B; Z). We then have the following result. Proposition 6.12 Let Mα denote the minimal model of Mα . Then

Mα (8) = ∧α , where Mα (8) is the degree 8 Postnikov piece of Mα .

6.2

Grove’s question

Proof The principal S1 -bundle S1 → Mα → B is modeled in a standard way by a relative model of the form (MB ⊗ ∧(t), D),

D|MB = dMB ,

D(t) = αx1 + x2 − x3 .

This model is, of course, non-minimal, but we can prove the proposition by giving a map ρ : ∧α → (MB ⊗ ∧(t), D) that induces an isomorphism on cohomology through degree 8 and an injection on degree 9 cohomology. With this in mind, define ρ by: ρ(x1 ) = x1 ,

ρ(x2 ) = x2 ,

ρ(y) = y,

ρ(zα ) = z + t (x33 + (αx1 + x2 )x23 + (αx1 + x2 )2 x3 + (αx1 + x2 )3 ). Direct calculation then shows that ρ satisfies the required properties.

Since there were an infinite number of distinct algebras ∧α , we have now obtained an infinite number of manifolds Mα with distinct rational homotopy types. To complete the proof of Theorem 6.9, we must show that these manifolds have metrics with the right curvature (and diameter) properties. This is where we shall use Theorem 6.8. In order to apply this result, we need to write the Mα as image spaces of submersions. With this in mind, define M to be the total space of a principal T 3 -bundle over B classified by x1 +x2 +x3 using the isomorphisms H 2 (B; Z) ∼ = [B, K(Z3 , 2)] ∼ = [B, BT 3 ]. Call this map ∼ =

f and note that f induces an isomorphism f# : π2 (B) → π2 (BT 3 ). Now let T 3 = {(z1 , z2 , z3 ) ∈ C | |zi | = 1} and write Tα2 for the subgroup generated by all (1, z, z) and (z, 1, zα ). The projection T 3 → T 3 /Tα2 induces pα : BT 3 → B(T 3 /Tα2 ) = BS1 . We then obtain the circle bundle S1 = T 3 /Tα2 → M/Tα2 → B with classifying map fα = pα ◦ f . Proposition 6.13 As circle bundles, Mα ∼ = M/Tα2 . Proof The Euler class of the bundle M/Tα2 → B is determined on H 2 , so we consider the dual map (fα )∗ : H2 (B; Z) → H2 (B(T 3 /Tα2 ); Z) = H2 (BS1 ; Z). Because the elements (1, z, z) and (z, 1, zα ) are killed in T 3 , on the homology level, we have (fα )∗ (0, 1, 1) = 0

and

(fα )∗ (1, 0, α) = 0.

Then, writing (fα )∗ (a, b, c) = pa+qb+rc (with (a, b, c) a coordinate expression with respect to the basis consisting of the homology duals of x1 , x2

247

248

6 : Curvature

and x3 respectively), the equations above give q + r = 0 and p + αr = 0. Therefore, (fα )∗ (a, b, c) = −αra − rb + rc = −r(αa + b − c). Since (fα )∗ is surjective, we must have r = ±1. Let ι¯ be the generator of H2 (BS1 ; Z) with ι the dual generator of H 2 (BS1 ; Z). The dual map (fα )∗ gives the Euler class eα by eα = (fα )∗ (ι) = Ax1 + Bx2 + Cx3 . We can compute as follows. A = (Ax1 + Bx2 + Cx3 )(1, 0, 0) = (fα )∗ (ι)(1, 0, 0) = ι(fα )∗ (1, 0, 0) = ι(α¯ι) = α. Similarly, we find that B = 1 and C = −1. Hence, eα = (fα )∗ (ι) = αx1 + x2 − x3 . Because Mα and M/Tα2 have the same Euler classes, we have Mα ∼ = M/Tα2 . Now note that Mα = M/Tα2 = N/Sα1 , where N = M/S01 and S01 ⊂ T 3 is the circle generated by (1, z, z). Let’s also use the following notation. Let 1 S1 denote the subgroup generated by (1, 1, z), Sα1 = Tα2 /S01 , S = q(S1 ) and 1

Sα = q(Sα1 ), where q : T 3 → T 2 = T 3 /S01 is the projection. Finally, denote 1

by sα : N → Mα and s : N → N/S the obvious projections. Since T 3 acts freely on M, note that T 2 acts freely on N. The final step in the construction of the examples of Fang and Rong is the following. Proposition 6.14 The Mα have metrics with a uniform bound on the absolute value of sectional curvature and diameter. Proof First note that, given a T 2 -invariant metric on N, we can choose 1 metrics on Mα and N/S so that sα and s are submersions. Then the diameter condition will be fulfilled since diam(Mα ) ≤ diam(N). Now, by the definitions above, we can see that a small segment of the 1 1 Sα -orbit through an x ∈ N converges locally to a small segment of the S orbit through x. But then we see that r-balls in the horizontal space of sα converge (uniformly in Tx (N)) to the r-ball in the horizontal space of s. This implies that we have convergence of the respective horizontal projections, hα → h, as in the hypothesis of Theorem 6.8. Therefore, we obtain uniform convergence secMα ((sα )∗ (hα (P))) → sec

N/S

1

(s∗ (h(P))).

This then provides the desired uniform bound on (absolute value of) sectional curvature (using Proposition 6.3).

6.2

Grove’s question

Proof of Theorem 6.9 By Lemma 6.10, Lemma 6.11, Proposition 6.12 and Proposition 6.14, the Mα (for all α large enough) provide an infinite number of mutually nonrationally homotopic manifolds in M≤D −1≤sec≤1 (n) once we recognize that the metric on N may be scaled to have −1 < secN < 1 while only increasing the diameter to some value D. Example 6.15 As we saw in Example 2.72 (also see the proof of Proposition 7.17), a relative extension of the form (where subscripts denote degrees) (∧(e2 ), d = 0) → (∧(e2 , x4 , y7 , z9 ), D) → (∧(x4 , y7 , z9 ), d) with differential given by choosing a ∈ Z+ (say), D(e2 ) = 0, D(x4 ) = 0, D(y7 ) = x24 + ae24 , D(z9 ) = e25 has a topological realization as a principal circle bundle M → Na → BS1 (where BS1 is modeled by (∧(e2 ), d = 0)) with M having the rational homotopy type of S4 × S9 and N being a smooth 12-manifold (see Theorem 3.2). It can easily be seen from the rational cohomology algebra that, for choices a1 a1 and a2 , Na1 has the same rational homotopy type as Na2 only if is a a2 rational square (see Example 2.38). Therefore, if we choose ai , i = 1, . . . , ∞ ai so that no is a square (e.g. ai is the ith prime), then the Nai give an infinite aj number of rational homotopy types. Note, however, that the real minimal models of all of these Nai are isomorphic since real square roots always exist for positive numbers. This brings up the following. Question 6.16 Can the Na be given metrics with secNa ≥ 0? Now, M Q S4 × S9 and S4 × S9 has non-negative sectional curvature, so if it were true that this condition propagates across rational equivalences, then M would have secM ≥ 0 too. Question 6.17 If two closed manifolds have the same (rational) homotopy type and one manifold has a metric of non-negative sectional curvature, does the other? On the other hand, we may ask whether Eschenburg’s method can produce the Na , thereby providing non-negative curvature.

6.2.2

Totaro’s approach

Now let’s consider the approach of B. Totaro [256]. Note that this approach allows the construction of counterexamples in the lowest possible dimension where they can exist, dimension 7 (see [259]). This is achieved at the

249

250

6 : Curvature

cost of using more complex algebra than that presented in Section 6.2.1. (Indeed, in [256], Totaro uses techniques from algebraic geometry to verify that he has an infinite number of mutually nonrationally equivalent six-dimensional biquotients with non-negative sectional curvature.) Theorem 6.18 ([256]) There exist real numbers C and D such that the subclass of M≤D −1≤sec≤C (7) consisting of simply connected manifolds contains infinitely many rational homotopy types. Proof First we define a commutative graded algebra by H = ∧(x0 , x1 , . . . , x4 )/R where xi , i = 0, . . . , 4, are of degree 2 and R is the ideal generated by x20 = x1 x2 , x21 = x2 x3 , x22 = x3 x4 , x23 = x4 x0 , x24 = x0 x1 and all the other products xi xj . One can easily see by direct computation that H 2 = Q[x0 , . . . , x4 ], H 4 = Q[x20 , . . . , x24 ] and H 6 = Q[x20 x1 = x21 x2 = x22 x3 = x23 x4 = x24 x5 ]. Notice that H is of finite total dimension and it satisfies Poincaré duality. Therefore, it can be realized by a manifold M of dimension 6 (see Theorem 3.2). /E / M determined We now construct a principal bundle (S1 )5 1 5 from the classifying map M → (BS ) defined by the cohomology classes x0 , . . . , x4 . Note that E is a manifold of dimension 11 endowed with a free action of (S1 )5 . We choose a Riemannian metric, invariant by the action of (S1 )5 , and we scale the metric to have curvature greater than −1. Denote by D the diameter of E. We now consider quotients of E by tori (S1 )4 ; these are simply connected manifolds Y of dimension 7 satisfying the following properties: • equipped with the induced metric, they are of diameter ≤ D; • by Proposition 6.6, we know that a submersion increases the sectional

curvature; therefore Y is of curvature ≥ −1;

• from Theorem 6.8, we know that, for the set of tori (S1 )4 ⊂ (S1 )5 , the

curvatures of the corresponding Y are uniformily bounded. This means that there exists a real number C such that the curvatures of all these manifolds Y are ≤ C. Let’s now study some of the spaces Y. We determine their rational homotopy type from a Sullivan model. Denote by MM a minimal model of M. Since p : Y → M is a S1 -principal bundle with base M, Y has a model of the following type:

MY = (MM ⊗ ∧x, d),

6.2

Grove’s question

where the differential d on MM coincides with the differential of 2 the min3 3 a 1 x3 imal model MM and dx ∈ M2M . We choose dx = a0 x0 − a1 x1 + a20 where a0 , a1 are rational numbers. From direct computations, we can check to H4 (M) has a kernel of dimenthat the multiplication by dx, from H 2 (M) sion 1 generated by z = a0 x0 + a1 x1 + sequence / H r (Y)

H r (M)

a20 a1

/ H r−1 (M)

x2 . From the Gysin exact

∪dx

/ H r+1 (M)

and the fact that dim H 2 (M) = dim H 4 (M) = 5, we see that Coker(∪ dx) also has dimension 1. Now, H 3 (M) = 0, so the Gysin sequence then says that H 4 (Y) has dimension 1. Clearly, we have that H 2 (Y) ∼ = H 2 (M)/(dx) has dimension equal to 4. Now, we have a commutative diagram p∗ ⊗p∗

/ H 2 (Y) ⊗ H 2 (Y) lll lll l l ll ψ l v ll

H 2 (M) ⊗ H 2 (M) ϕ

H 6 (M) ∼ =Q

where the top arrow is clearly surjective (since p∗ : H 2 (M) → H 2 (Y) is surjective) and the homomorphism ϕ is the cup-product with the class z, ϕ(α ⊗ β) = α ∪ β ∪ z. Since dx ∪ z = 0, ϕ induces a quadratic form ψ on H 2 (Y). From the definition of ϕ as a cup-product, we deduce its matrix on the five-dimensional space H 2 (M), ⎛ ⎜ ⎜ ⎜ ⎜ ⎝

a1 a0 0 0 0

a0 a2 a1 0 0

0 a1 0 a2 0

0 0 a2 0 0

0 0 0 0 a0

⎞ ⎟ ⎟ ⎟ ⎟ ⎠

where a2 = a20 /a1 . The 4 × 4 matrix in the right bottom corner is a matrix a7 for ψ; its determinant is − 03 . Note that we are looking for isomorphism a1 types so we have to admit a multiplication of the quadratic form ψ by a scalar. Since dim H 2 (Y) = 4, which is even, the determinant in Q/(Q∗ )2 of the quadratic form is unchanged. Therefore, the determinant is an invariant of isomorphism classes only as an element in Q/(Q∗ )2 . The determinant of

251

252

6 : Curvature

ψ gives us an element −

a0 ∈ Q/(Q∗ )2 . a1

Finally, note that all values of Q/(Q∗ )2 are reached for appropriate choice of (a0 , a1 ) ∈ Q × Q. Therefore we get an infinite number of different rational homotopy types corresponding to the spaces Y. Remark 6.19 Totaro’s examples above are biquotients and this is no accident. In fact, almost all known closed manifolds having non-negative sectional curvature are biquotients. The only such manifolds which are not defined as biquotients are the Cheeger manifolds obtained as the connected sums of two spaces from the list: RP(n), CP(n), HP(n) and CayP (where the last is the Cayley plane) and an example of Grove and Ziller. This is discussed in [255], where Totaro also determines the overlap; that is, he classifies which Cheeger manifolds are diffeomorphic to biquotients and which are not. For instance, CP(4)#HP(2) is not even homotopy equivalent to any biquotient. Furthermore, in contrast to Theorems 6.18 and 6.9, Totaro proves in [255] that there are only finitely many diffeomorphism classes of 2-connected biquotients of a given dimension.

6.3

Vampiric vector bundles

In [56], the following theorem was proved. Theorem 6.20 (The soul theorem) Let M be a complete noncompact Riemannian manifold with non-negative sectional curvature. Then there i

exists a totally geodesic and totally convex compact submanifold S → M such that M is diffeomorphic to the normal bundle of the embedding i. Remark 6.21 (Cheeger–Gromoll splitting) In [55], it was shown that a man that splits ifold M with non-negative Ricci curvature has a universal cover M n n isometrically as M = M0 × R , with R flat and M0 compact. See Exercise 6.1 for the definition and properties of Ricci curvature. We shall only use this result peripherally, so we simply mention it in this remark. We do note, however, that the splitting comes from peeling off “lines” from M. Recall that a line is a geodesic that minimizes distance between each pair of points on it. The submanifold S is called the soul of M. Recall that S is totally geodesic in M if every geodesic in M starting from a point of S in a direction tangent

6.3


to S is in fact a geodesic in S. The submanifold S is totally convex if, for any two points of S, any geodesic in M joining them (which exists since M is complete) lies in S. Remark 6.22 Right away, note some immediate consequences of the soul theorems 6.20. 1. Because S is totally geodesic, S also has non-negative sectional curvature. 2. The inclusion i : S → M is a homotopy equivalence. Therefore, if M is contractible (e.g. M = Rn ), then S = ∗. is contractible and also 3. If Mn = K(π , 1) is a closed manifold, then M has non-negative sectional curvature. By Cheeger–Gromoll splitting, we ∗, so M0 = ∗ as well and, thus, M = Rn = M0 × Rn . But M have M isometrically. Since covering transformations act as isometries, we then see that M = Rn /π is flat (see Example 6.4). 4. Given any closed manifold S with non-negative sectional curvature, it is a soul; namely, S is the soul of S × Rn . Note that we may consider the product S × Rn as the trivial Rn -vector bundle over S. The last item elicits the following question. Question 6.23 Given a vector bundle Rn → E → M, where M is a closed manifold with non-negative sectional curvature, does E admit a metric with non-negative sectional curvature? Recently, algebraic models have been used to give obstructions to the existence of such a metric on the total space of a vector bundle (see [22] and [23]). This will be the focus of this section, but in order to understand the whole situation better, we will first give the original negative examples for Question 6.23 obtained in [219]. Since these examples don’t have souls, we refer to them as vampiric vector bundles.

6.3.1

The examples of Özaydin and Walschap

The basis of the Özaydin–Walschap approach to constructing the examples mentioned above is the following result. Theorem 6.24 Let ξ : Rk → E → Mn be a vector bundle over a compact, flat manifold M. The following are equivalent: 1. E admits a complete metric with secE ≥ 0. 2. E admits a complete metric with secE = 0. 3. E is diffeomorphic to the total space of a rank k vector bundle over M that admits a flat Riemannian connection.

253

254

6 : Curvature

4. E ∼ =diffeo Rn ×π1 M Rk , where π1 (M) acts on Rn by covering transformations (as in Example 6.4) and on Rk by some orthogonal representation. We can generalize the notion of flat manifold to bundles as follows. Definition 6.25 A principal G-bundle ξ : G → P → X is flat if the classifying map X → BG factors up to homotopy as X C

C

C

/ BG < y i yy y yy yy

C!

BGd where Gd is the Lie group G taken with the discrete topology and i : Gd → G is the canonical continuous identification of Gd and G as sets. See Remark 6.27 for an explanation of the use of the term “flat” and its relation with flat manifolds. The following result gives an equivalent criterion for flatness that leads to better interpretations. Proposition 6.26 The principal G-bundle ξ : G → P → X is flat if and only if there is a factorization up to homotopy / BG X E EE ρ w; EE w EE w E" w Bπ1 X

where Bπ1 X → BG is induced by a homomorphism π1 X → G and ρ : X → Bπ1 X = K(π1 X, 1) is the “canonical” inclusion obtained by attaching cells of dimension 3 and greater to kill all homotopy groups above degree 1. Proof Suppose the homomorphism π1 X → G exists which provides the factorization above. Since π1 X is a discrete group, there is then a factorization π1 X

D

D

D

/ G > ~ ~~ ~ ~~ ~~

D!

Gd

6.3


which induces the desired homotopy factorization / BG ; O v vv v ρ v i vv vv / BGd Bπ1 X

X

Now assume that X → BG factors through BGd . The induced map of fundamental groups is then π1 X → Gd and this, in turn, induces Bπ1 X → BGd . Now, the homotopy class of a map X → K(Gd , 1) = BGd is determined by the induced homomorphism on fundamental groups. Thus we obtain a homotopy commutative diagram / BG X H O HH HH HH i ρ HH # / BGd Bπ1 X

Remark 6.27 The principal bundle induced from the canonical map X → → X. The effect Bπ1 X is simply the universal covering fibration π1 X → X of Proposition 6.26 is to reduce the group of the principal bundle ξ to ×π X G, where π1 X acts on the fundamental group π1 X; hence, P ∼ = X 1 X by covering transformations and on G via the homomorphism of the proposition and multiplication in G. A vector bundle Rk → E → X is flat if the associated principal O(k)-bundle O(k) → P → X is flat and the consequent homomorphism ρ : π1 X → O(k) gives ×π X R k . E = P ×O(k) Rk = X 1 Of course, the bundle is called flat because it has a flat Riemannian connection (see [194] and [198, Appendix C]). A Riemannian connection ∇ for a vector bundle is the analogue of the covariant derivative on vector fields. The analogue of a vector field is a section of the bundle and the connection “differentiates” the section subject to properties analogous to those in Proposition 6.1. The connection ∇ is flat if, in the neighborhood of each point of the base X, there is a basis of sections s1 , . . . , sk with ∇(si ) = 0 for i = 1, . . . , k. This is actually a convenient equivalent condition to the vanishing of the bundle’s curvature (see [28, Section 5.3, Theorem 5]). But now consider the change of basis ti = j aji sj when moving to a new chart

255

256

6 : Curvature

with local basis t1 , . . . , tk . The properties of ∇ (including the hypothesized flatness) give 0 = ∇(ti ) = (daji sj + aji ∇(sj )) = daji sj . j

j

Hence, daji = 0 for all the aji . These transition functions are therefore locally constant and factor through the structure group made discrete. Since the transition functions are the same for the associated principal bundle, we see that this notion of flatness corresponds to the principal bundle notion. If the flat vector bundle is the tangent bundle Rn → TX → X n , then the flat Riemannian connection is precisely the connection on X associated to the metric. Hence, X is a flat manifold. This is the “connection” between flat manifolds and flat vector bundles. Now, in the situation of (3) and (4) of Theorem 6.24, since X is a flat manifold, the Bieberbach theorem tells us = Rn . Hence, the total space E of the flat vector bundle is that X ×π X Rk = Rn ×π X Rk . E = P ×O(k) Rk = X 1 1 Proof of Theorem 6.24 The discussion above shows the equivalence of (3) and (4). Also, (2) clearly implies (1) while (4) implies (2) because π1 M acts on Rn × Rk = Rn+k by isometries (and so the zero curvature of Rn+k is preserved under the quotient operation E = (Rn × Rk )/π1 M). Now assume (1) and let S be a soul of E. Then secS ≥ 0 and S E M = K(π , 1); therefore, S is flat by Remark 6.22 (3). Now, (see Example 6.4) flat manifolds are classified up to diffeomorphism by their fundamental groups, so S is, in fact, diffeomorphic to M. Now let p : E → E denote the universal E be the restriction to the (Riemannian) covering of E and let S = p−1 (S) ⊂ soul. The pullback diagram

S

S

/ E

p

/ E

shows that S → S is a fibration. Moreover, because S → E is a homotopy equivalence, it induces an isomorphism of fundamental groups and this means that the map S → S is the universal (Riemannian) covering. Since S = K(π, 1), S is contractible; furthermore, by the Cheeger–Gromoll splitting S is theorem (Remark 6.21), we see that S ∼ =isom Rn . Now observe that totally convex in E. This follows because the geodesic in E joining two points in S pushes down via p to a geodesic in E joining two points in S. Then, since S is totally geodesic in E, the whole geodesic must lie in S. Thus,

6.3


the original geodesic lies in p−1 (S) = S. This has the effect that a line (see Remark 6.21) in S is a line in E. Hence, the Cheeger–Gromoll splitting of E has the form E∼ S. =isom E0 × Rn = E0 × Then we see that E splits locally isometrically as a product over S. Therefore, the Riemannian connection on the normal bundle of the inclusion S → E (induced from the Riemannian connection on E) is flat. Since E is diffeomorphic to the total space of this normal bundle (by the Soul theorem), we have proven (3). With these preliminaries out of the way, we can now state the main result of [219]. Theorem 6.28 Let ξ : R2 → E → Mn be an oriented vector bundle, where Mn is a compact flat manifold. Then E admits a complete metric of nonnegative curvature if and only if the rational Euler class e(ξ )Q is zero. In order to prove the theorem, we need a lemma. Lemma 6.29 Let ξ : Rk → E → M be an oriented vector bundle over a compact oriented manifold M. If E ∼ =homeo F × R, then eZ (ξ ) = 0, where eZ (ξ ) is the integral Euler class of ξ . ∼ =

Proof Let h : E → F × R be the given homeomorphism. Consider the composition s

pr

h

it

h−1

M → E → F × R → F → F × {t} → F × R → E, which we denote by φ. Here, s is the zero section and t is larger than any t such that (f , t ) ∈ Im(h ◦ s). This can be done since M is compact. Now, the pr

it

composition F × R → F → F × R is homotopic to idF×R by H(f , r, u) = (f , r(1 − u) + ut). Thus, φ h−1 ◦ idF×R ◦ h ◦ s s. Now, the (integral) Euler class is defined by s∗ () = e(ξ ), where ∈ k H (E, E/s(M)) is the Thom class and is its restriction in H k (E). But, since φ s, we have φ ∗ () = e(ξ ). However, φ(M) ∩ s(M) = ∅ because of the map it . Hence, φ factors as φ

/ E M G x; GG x x GG xx GG GG xx

x ¯ φ -x # E/s(M)

257

258

6 : Curvature

We then have the commutative diagram / H k (E)

H k (E, E/s(M))

φ∗

/ H k (E/s(M)) rr rrr r r xrrr φ¯ ∗

H k (M) with the Euler class computed two ways: once with the definition and secondly using the factorization of φ given above.

/ _ φ∗

e(ξ )

/ =0 } } }} }} ¯ ∗ } ~} φ

Hence e(ξ ) = 0 since the composition H k (E, E/s(M)) → H k (E) → H k (E/s(M)) is zero by exactness. Proof of Theorem 6.28 First, let e(ξ )Q = 0. By Corollary 6.48, the associated principal S1 -bundle is flat. Therefore, the vector bundle is flat and from the equivalence of (3) and (1) in Theorem 6.24, we obtain a non-negatively curved metric on E. Now suppose secE ≥ 0. Because Mn is flat, there exists a finite cover f : T n → Mn . Take the pullback bundle / E

E(f ∗ ξ )

Tn

f

/ Mn

and a pullback metric on E(f ∗ ξ ) with sec ≥ 0. Now, a finite cover induces an injection in rational cohomology, f ∗ : H ∗ (M; Q) → H ∗ (T n ; Q). (Therefore, if e(ξ )Q were nonzero, then e(f ∗ (ξ ))Q would also be non-zero.) By the equivalence of (3) and (1) in Theorem 6.24, E(f ∗ ξ ) is diffeomorphic to the total space of a flat R2 -bundle η over T n . But then we know that e(η)Q = 0 and, since H 2 (T n ; Z) is torsion-free, in fact e(η)Z = 0 as well. Of course, the associated principal S1 -bundle is classified by e(η)Z , so η is trivial: E(η) = T n × R2 . But then we have E(f ∗ ξ ) ∼ = (T n × R) × R. =diffeo E(η) = T n × R2 ∼ By Lemma 6.29, we have e(f ∗ ξ ) = 0 and, therefore, e(ξ )Q = 0.

6.3


Example 6.30 We can now apply Theorem 6.28 to give explicit examples of vector bundles over compact non-negatively curved manifolds whose total spaces have no metrics of non-negative curvature. Oriented R2 -bundles over T 2 are classified by homotopy classes of maps, [T 2 , BSO(2)] = [T 2 , BS1 ] = [T 2 , K(Z, 2)] = H 2 (T 2 ; Z) = Z. Explicitly, each k ∈ Z corresponds to an Euler class e(ξk ). Thus, since the rational Euler class must vanish in order for the total space of the bundle to have non-negative curvature, we have the following. Theorem 6.31 Only the total space of the trivial R2 -bundle over T 2 possesses a metric with non-negative curvature.2

6.3.2

The method of Belegradek and Kapovitch

Now that we have seen the particular examples of Özaydin and Walschap, we can explore more general (necessary) conditions for the existence of total spaces of bundles with metrics of non-negative curvature. In the following, we shall mainly follow [23], but also see [22] and [220]. We shall see that a main tool in the understanding of this problem is the most fundamental algebraic model of all, rational cohomology. First, the following result of [22] generalizes Theorem 6.20 and Remark 6.21 and, since we are attempting to imitate the normal bundle situation, focusses our attention on bundles over products C × T k . Theorem 6.32 Let M be a complete manifold with secM ≥ 0. There exists a finite cover M → M such that M ∼ =diffeo N ×T where N is a complete simply connected open manifold with secN ≥ 0 and T is a torus. Furthermore, if S is a soul of M, then the diffeomorphism carries S to C × T where C is a soul of N. Remark 6.33 Note that, since C is a soul of N (i.e. N is diffeomorphic to the normal bundle of the inclusion C → N), the theorem is saying that M is, in fact, a bundle η over C × T of the form η = ξC × T, where ξC is a bundle over C. With this theorem in mind, let’s make the following. Definition 6.34 Suppose C is a closed simply connected manifold, T is a torus and k ∈ Z. The triple (C, T, k) is said to be splitting rigid if, given a rank k vector bundle ξ : E(ξ ) → C × T with secE(ξ ) ≥ 0, then there exists 2 So the other bundle total spaces do not have non-negative curvature and, hence, have no souls. That is, they are vampires!

259

260

6 : Curvature

a finite covering map p : T → T such that (idC × p)∗ (ξ ) ∼ = ξC × T, where ξC is a rank k bundle over C. This definition can be used to obtain an obstruction (i.e. a necessary condition) for the existence of a metric of non-negative sectional curvature on the total space of a bundle. Namely, since most bundles over C × T do not split after passing to a finite cover, the following result says that any such non-splitting bundle over the product of a biquotient and a torus has a total space which does not have sec ≥ 0. Theorem 6.35 ([23]) Let G//H be a simply connected biquotient, where G is compact connected and H is semisimple and connected. Then, for any torus T and k ≤ 4, (G//H, T, k) is splitting rigid. Let’s begin to understand this result (with the proof following later). Denote by Char(C, k) the following characteristic algebra inside H ∗ (C; Q).

Char(C, k) =

⎧0[m/2] 4i ⎪ ⎨ i=1 H (C; Q) ⎪ ⎩0[m/2] i=1

H 4i (C; Q) ⊕ H 2m (C; Q)

if k = 2m + 1 if k = 2m.

Here, [−] denotes the greatest integer function. The important point is that, for a rank k bundle over C, Char(C, k) contains the Pontryagin classes when k is odd and the Pontryagin classes and Euler class when k is even. (We refer to these classes as the rational characteristic classes of a bundle.) This is important because of the following result. Theorem 6.36 ([22]) Let ξ and η be vector bundles over C × T of the same rank. If ξ and η have the same rational characteristic classes and ξ |C ∼ = η|C , ∗ ∼ then there exists a finite covering f : C × T → C × T such that f ξ = f ∗ η. Now, given a bundle ξ over C × T whose rational characteristic classes lie in Char(C, k), clearly the rational characteristic classes of ξ and ξ |C × T are the same and, by definition ξ |C = ξ |C . Therefore, Theorem 6.36 implies that there is a finite covering f : C × T → C × T with f ∗ ξ ∼ = f ∗ (ξ |C × T). But since C is simply connected, the finite covering has the form idC × f˜ ; thus, f ∗ (ξ |C × T) = ξ |C × T. This means that we have the following result. Lemma 6.37 A bundle ξ over C × T (as above) splits if and only if the rational characteristic classes lie in Char(C, k). We will now use this result to obtain a criterion for splitting rigidity.

6.3


Proposition 6.38 ([23]) If every self-homotopy equivalence of C × T preserves the characteristic algebra Char(C, k) in H ∗ (C × T; Q) ∼ = H ∗ (C; Q) ⊗ ∗ H (T; Q), then (C, T, k) is splitting rigid. Proof Suppose ξ is a vector bundle over C × T with secE(ξ ) ≥ 0. By Remark 6.33, there is a finite cover E → E(ξ ) with E ∼ = N ×T , where N is simply connected and secN ≥ 0. Since N is simply connected, we see in fact that T = T. Also, for S the soul of E , we have S ∼ = CN × T where CN is a ∼ soul of N, so N = E(ζ ), where ζ is the normal bundle of CN → N. Therefore, without loss of generality, we take E(ξ ) ∼ = E(η) = E(ζ )×T, where ζ is a vector bundle over a simply connected manifold C . Thus, ξ and η are two different vector bundle structures on the same manifold X = E(ξ ) = E(η). The respective zero sections C × T → E(ξ ) = X, C × T → E(η) = X identify C × T, C × T as submanifolds of X and we consider the bundles ξ and η as the normal bundles of these respective inclusions. Note that the composition of the first zero section map with the projection of the bundle η gives a homotopy equivalence g : C × T → C × T which can be taken to have degree one by orienting X properly. Furthermore, the obvious maps gC : C → C , gT : T → T are homotopy equivalences, so we may form the −1 self-homotopy equivalence h = (gC × gT−1 ) ◦ g : C × T → C × T. The inclusions and projections give maps h

hC : C → C × T → C × T → C h

hT : T → C × T → C × T → T. These are, in fact, homotopic to the respective identity maps. For instance, if the inclusions and projections for C, C are denoted by iC , pC , iC , pC respectively, then we have −1 −1 −1 × gT−1 )giC gC pC giC gC gc idC hC = pC hiC = pC (gC −1 −1 since pC ◦ (gC × gT−1 ) gC ◦ pC and pC ◦ g ◦ iC = gC . By assumption ∗ h maps Char(C, k) to itself, so it is plain that g ∗ maps Char(C , k) to Char(C, k). Since ξ and η are normal bundles to the respective inclusions of the zero sections, we know by Poincaré duality that the Euler classes obey e(ξ ) = [C × T], e(η) = [C × T]. Therefore, since deg(g) = 1, for any class y ∈ H∗ (C × T), we obtain

g ∗ (e(η)), y = e(η), g∗ (y) = [C × T], g∗ (y) = g ∗ ([C × T]), y = [C × T], y = e(ξ ), y. So we see that g ∗ (e(η)) = e(ξ ). Now, e(η) ∈ Char(C , k) since η ∼ = ζ ×T and g ∗ maps Char(C , k) to Char(C, k), so e(ξ ) ∈ Char(C, k).

261

262

6 : Curvature

Now let’s consider Pontryagin classes. Since g : C × T → C × T is defined by zero

πη

C × T −→ X → C × T and πη is a homotopy inverse to the η-zero section C × T → X, we see that g, thought of as a map g : C × T → X is homotopic to the zero section inclusion C × T → X. Then we obtain g ∗ (TX|C ×T ) = TX|C×T . Now, because we view ξ as the normal bundle to the inclusion C × T → X, and we may apply the Whitney sum formula for the total Pontryagin class p to the bundle TX|C×T = T(C × T) ⊕ ξ to obtain p(T(C × T))p(ξ ) = p(T(C × T) ⊕ ξ ) = p(TX|C×T ) = p(g ∗ (TX|C ×T )) = g ∗ (p(TX|C ×T )) = g ∗ (p(T(C × T) ⊕ η)) = g ∗ (p(T(C × T))g ∗ (p(η)). Now, T is parallelizable, so p(T(C×T)) = p(TC) ∈ H ∗ (C) and p(T(C × T)) = p(TC ) ∈ H ∗ (C ). Also, p(T(C × T)) is a unit in H ∗ (C), so we can write p(ξ ) = p(T(C × T))−1 g ∗ (p(T(C × T))g ∗ (p(η)). Now p(η) ∈ H ∗ (C ) since η ∼ = ζ × T and g ∗ (Char(C , k)) ⊆ Char(C, k), so ∗ each factor of p(ξ ) lies in H (C). Hence, p(ξ ) ∈ Char(C, k) as well. Because the rational characteristic classes are in Char(C, k), by the discussion following Theorem 6.36, we see that (C, T, k) is splitting rigid. Remark 6.39 Note that the proof shows that, without loss of generality, we can always assume that the restriction to C of the self-homotopy equivalence is homotopic to idC . The link between self-homotopy equivalences of C×T and algebraic operations on cohomology is provided by the the structure of H ∗ (T) (where we will understand coefficients to be Q from this point forward). The torus cohomology H ∗ (T) is an exterior algebra ∧(e1 , . . . , es ) with vector space basis t0 = 1, t1 = e1 , . . . , ts = es , ts+1 = e1 e2 etc. (in some chosen, perhaps lexicographic, ordering). Now, H ∗ (C × T) ∼ = H ∗ (C) ⊗ H ∗ (T), so a homotopy equivalence (with h|C = idC by Remark 6.39) h : C × T → C × T

6.3


induces an automorphism h∗ : H ∗ (C × T) → H ∗ (C × T) such that, for c ∈ H ∗ (C) ⊂ H ∗ (C × T), δi (c)ti h∗ (c) = c + i

where we have not written the ⊗ in the free H ∗ (C)-module H ∗ (C×T). It can be shown that δ1 is a derivation and, further, the fact that t12 = 0 shows that the mapping φ1 : H ∗ (C × T) → H ∗ (C × T) defined by φ1 (ct) = ct − δ1 (c)t1 t is a graded algebra isomorphism (with inverse φ1−1 (ct) = ct+δ1 (c)t1 t). Then δi (c)ti ) φ1 (h∗ (c)) = φ1 (c + i≥1

= c − δ1 (c)t1 + δ1 (c)t1 + =c+

φ1i (c)ti

i≥2

φ1i (c)ti .

i≥2

So we see that φ1 (h∗ (c)) resembles the identity with error terms involving only the ti with i ≥ 2. The idea is now to continue this process until the error terms are zero. We then shall have written h∗ in a very precise way that will permit a connection with derivations of H ∗ (C). Now consider the formula for φ1 (h∗ (c)). As usual, the first perturbation term φ12 is a derivation, so φ2 (ct) = ct − φ12 (c)t2 t is an automorphism of H ∗ (C × T) with φ2 (φ1 (h∗ (c))) = c + φ2i (c)ti . i≥3

We continue in this fashion to obtain automorphisms φk for k = 1, . . . ,

k (c)t and such that N = ks with φk (c) = c − φk−1 k (φN ◦ . . . ◦ φ1 ◦ h∗ )(c) = c. Then the isomorphism h∗ evaluated on c ∈ H ∗ (C) may be written as −1 ◦ . . . ◦ φ1−1 )(c) h∗ (c) = (φN k (c)t . where φk−1 (c) = c + φk−1 k

Theorem 6.40 If every negative degree derivation of H ∗ (C) vanishes on the characteristic algebra Char(C, k), then (C, T, k) is splitting rigid for each T. Proof By Proposition 6.38, we must show that every self-homotopy equivalence of C × T preserves Char(C, k). As noted in Remark 6.39, we may

263

264

6 : Curvature

assume the restriction of the equivalence to C is idC . With this in mind, the discussion above pertains and the induced isomorphism on cohomology has the form −1 ◦ . . . ◦ φ1−1 )(c) h∗ (c) = (φN k (c)t . But each φ k (c) = 0 by the hypothesis where φk−1 (c) = c + φk−1 k k−1 of the theorem since c ∈ Char(C, k). Thus, h∗ (c) = c, so h∗ preserves the characteristic algebra Char(C, k).

Corollary 6.41 (also see [220]) If H ∗ (C) has no nonzero negative degree derivations, then (C, T, k) is splitting rigid for each T and k. In particular, by Theorem 4.36 and [237], we know that simply connected Kähler manifolds and maximal rank homogeneous spaces have cohomology algebras admitting no negative degree derivations. Therefore, such manifolds are prime candidates for C’s. As we have seen, splitting rigidity is an obstruction to the total space of a bundle over C × T having non-negative sectional curvature; so now we see further that the question of the existence of metrics with non-negative curvature is intimately involved with the algebraic structure of H ∗ (C). Finally, let’s use Theorem 6.40 to prove Theorem 6.35. Proof of Theorem 6.35 The group H is semisimple, so the fibration G → G//H → BH shows that G is also semisimple and, since π2 (G) = 0 and π2 (BH) = π1 (H) is finite, π2 (G//H) is finite. Therefore, over the rationals, G//H is 2-connected. Therefore, the minimal model MG//H = (∧V, d) of G//H has V 1 = V 2 = 0. In Corollary 3.51, we saw that G//H has a pure minimal model, so the generators in V 3 appear in no differential. Hence, d), where V 3 = 0. by Theorem 2.77, MG//H = (∧V 3 , d = 0) ⊗ (∧V, 2 Now, because H (G//H) = 0, by Theorem 6.40 we need only consider rank 3 and rank 4 bundles over G//H × T. Now, by definition, d), so we only Char(G//H, 3) = Char(G//H, 4) = H 4 (G//H) = H 4 (∧V, have to consider derivations of degree −1 and degree −4. Suppose θ is a derivation of degree −4 and let c ∈ H 4 (G//H). Since G//H is finite dimensional, there is some m ∈ Z+ , such that cm = 0 and cm+1 = 0. Then 0 = θ(cm+1 ) = (m + 1)θ (c)cm and, since θ (c) ∈ Q and cm = 0, it must be true that θ (c) = 0. Thus θ vanishes on H 4 (G//H).

6.4

Final thoughts

Now suppose that θ is a derivation of degree −1 and let c ∈ H 4 (G//H). Then θ (c) = x ∈ V 3 . Again find m so that cm = 0 and cm+1 = 0. Then 0 = θ(cm+1 ) = (m + 1)θ(c)cm = (m + 1)xcm . d)) = 0 for Note that, if dim (G//H) = n and dim (V 3 ) = k, then H j (∧V, m 3 j > n − 3k. Since c = 0, we have 4m ≤ n − 3k. Also, x ∈ V , so 4m + 3 ≤ n−3k+3 = n−3(k−1) and, since k ≥ 1, 4m+3 < n. Therefore xcm is below the top dimension in cohomology. Because H ∗ (∧V, d) ∼ = H ∗ (∧V 3 , d = 0) ∗ ∗ 3 m ∗ ⊗ H (∧V, d), for x ∈ H (∧V , d = 0) and c ∈ H ((∧V, d)), we can only have xcm = 0 if x = 0 or cm = 0. Since cm = 0, we must have θ (c) = x = 0. Thus, θ vanishes on H 4 (G//H). Therefore, we see that all negative degree derivations vanish on Char(G//H, k) for k = 3, 4 and by Theorem 6.40, this means that, for k = 3, 4, (G//H, T, k) is splitting rigid for any T. Finally, we mention a result of [23] showing that not all spaces give splitting rigid triples. The proof of the result relies essentially on the properties of the minimal model of SU(6)/(SU(3) × SU(3)) (which is an example of a nonformal homogeneous space). Theorem 6.42 Let C = SU(6)/(SU(3) × SU(3)) and take dim T ≥ 2. Then there exists a rank 6 vector bundle which does not split, but whose total space has non-negative sectional curvature.

6.4

Final thoughts

Manifolds with sec ≥ 0 enjoy many special properties and are conjectured to have many more. For instance, Synge’s famous theorem says that a closed manifold M with secM > 0 is orientable if dim (M) is odd and is simply connected if dim (M) is even and M is orientable. Of course RP(2n + 1) is a prime example of the first case, while RP(2n) shows that the orientability hypothesis is necessary for the second case. Many years ago, Hopf conjectured that a closed manifold M with dim (M) even and secM ≥ 0 must have χ (M) ≥ 0. More recently, Gromov asked whether a closed manifold M with secM ≥ 0 must have i bi (M) ≤ 2dim (M) , where bi (M) is the ith Betti number of M. Recall from Theorem 2.75 that these assertions are indeed true for rationally elliptic manifolds (i.e. those with dim (π∗ (M)⊗ Q) < ∞). This then presents a tantalizing possibility which is enunciated in a far-reaching conjecture generally attributed to Raoul Bott. Conjecture 6.43 If M is a closed manifold with secM ≥ 0, then M is rationally elliptic.

265

266

6 : Curvature

The idea that a geometric constraint imposes conditions on the underlying algebraic topology of a manifold is not a new one, of course. We have already mentioned Hadamard’s theorem that a manifold with nonpositive sectional curvature must be a K(π , 1) and Bieberbach’s theorem characterizing flat manifolds. There is also Bochner’s famous result that a closed manifold of non-negative Ricci curvature must have first Betti number bounded above by the manifold’s dimension. The same hypothesis yields a similar restriction on the growth of the fundamental group (see [196]). In a different direction, we have seen in Section 5.7 that qualities associated with geometric entities such as the geodesic flow can radically limit possibilities for homotopy type as well. In this chapter, we have seen how algebraic models can be associated to geometric objects and how they can be used to construct geometric entities. Algebraic models are important ingredients in the solutions of geometric problems, so they have proven to be part of the tradition of obtaining geometric information homotopically. Henri Poincaré had a criterion for significance that he claimed for the mathematics that leads to a mathematical law. The beautiful symbiosis between homotopical algebra and geometry exemplifies this criterion and in Poincaré’s words, “Ce sont ceux qui nous révèlent des parentés insoupçonnées entre d’autres faits, connus depuis longtemps, mais qu’on croyait à tort étrangers les uns aux autres.” [“They reveal the kinship between other facts, long known, but wrongly believed to be strangers to one another.”]

6.5

Appendix

In this appendix, we will show that the rational Euler class for a principal circle bundle vanishes exactly when the bundle is flat (see [217]). Let Gd denote the Lie group G with the discrete topology and similarly for the and G d . universal covering group G → G Proposition 6.44 Let G be a connected Lie group and let π : G denote the universal covering. Then Gd is a pullback of π along the map i : Gd → G. discrete unravels all of the topology linking Proof Note first that making G the fibers of the universal covering of G. Now, for discrete spaces, only d = Gd × π1 (G). cardinality matters, so we have G Let P denote the pullback of π along the map i : Gd → G. Now, → G is a principal π1 (G)-bundle, so its pullback along i is one π: G → G is surjective, there is a set map s : G → G with also. Because π : G π ◦ s = idG . But Gd has the discrete topology, so s considered as a map

6.5

Appendix

is continuous. A principal bundle with section is trivial, so we s : Gd → G d . obtain P = Gd × π1 (G) = G Now let G = S1 with universal cover R and note that BS1 = K(Z, 2), BSd1 = K(Sd1 , 1) and BRd = K(Rd , 1). Consider the following pullback diagram.

Rd

Sd1

i

/ R π

i

/ S1

Of course the homomorphisms of groups shown are the usual ones making the sequence Z → R → S1 exact. Note that the pullback property guarantees that the fiber of Rd → Sd1 is Z included in Rd in the standard way and that the restriction of i to fibers is the identity. But now we notice an amazing fact. Because R → S1 is a fibration with R contractible, Rd is the homotopy fiber of Sd1 → S1 . Therefore, i : Sd1 → S1 classifies Rd → Sd1 and BSd1 → BS1 classifies BRd → BSd1 in the sequence of classifying spaces BZ

/ BR d

K(Z, 1)

/ K(Rd , 1)

/ BS1 d

/ BS1

/ K(S1 , 1) d

(6.1)

/ K(Z, 2)

Remark 6.45 Note that the first two maps are determined by the corresponding homomorphisms of groups. The last map also comes from the sequence of discrete groups; it is the class in Ext(Sd1 , Z) ⊆ H 2 (Sd1 ; Z) classifying the central extension Z → Rd → Sd1 (see [137]). Proposition 6.46 There exists a map φ : K(Z, 2) → K(Rd , 2) so that the sequence of fibrations (6.1) may be extended to a sequence of fibrations φ

K(Z, 1) → K(Rd , 1) → K(Sd1 , 1) → K(Z, 2) → K(Rd , 2). Proof This is, in fact, a standard result in topology (see [139, Theorem 7.1]). Let φ correspond to the usual inclusion of the integers into the reals, Z → Rd . Of course, this is the induced homomorphism at the beginning of the sequence K(Z, 1) → K(Rd , 1) as well. To see that we get a fibration on the right end, let F denote the homotopy fiber of φ. We must show

267

268

6 : Curvature

that F → K(Z, 2) is precisely the map K(Sd1 , 1) → K(Z, 2) in the original fibration sequence corresponding to the group extension Z → Rd → Sd1 as discussed in Remark 6.45. The long exact sequence in homotopy associated to the fibration shows immediately that F K(Sd1 , 1) since (because Z → Rd is injective) the only nontrivial homotopy group appears in degree one and is a quotient of the standard inclusion Z → Rd (which defines φ). We now extend to a Puppe sequence (leaving off the K(Rd , 2) on the right), K(Z, 1) → K(Rd , 1) → F = K(Sd1 , 1) → K(Z, 2), where we can see that the homomorphisms corresponding to the first two maps are the usual ones. The first is just a loop of φ, so that is clear. The second map is just a spatial realization of the connecting homomorphism in the homotopy sequence by K(Rd , 1) = K(Rd , 2) → K(Sd1 , 1). Since the connecting homomorphism was the usual group projection onto the quotient, the second map is the usual one too. The last map F → K(Z, 2) is a classifying map for the fibration K(Z, 1) → K(Rd , 1) → F = K(Sd1 , 1) since the Puppe sequence is constructed by taking consecutive homotopy fibers. But the fibration is determined by the group extension Z → Rd → Sd1 since the spaces are Eilenberg–Mac Lane spaces in degree one. Indeed, the classifying map corresponds to an element in Ext(Sd1 , Z) ⊆ H 2 (Sd1 ; Z) classifying the group extension. But the group extension is the usual one, so the element, and hence the classifying map F → K(Z, 2) are the usual ones as well. A principal S1 -bundle ξ : S1 → P → X is classified by a map κ : X → BS1 = K(Z, 2). This map corresponds to a degree 2 cohomology class by the standard identification of homotopy classes [X, K(Z, n)] with cohomology H n (X; Z). In fact, H 2 (K(Z, 2); Z) ∼ = Hom(Z, Z), so there is an element ι ∈ H 2 (K(Z, 2); Z) corresponding to the identity homomorphism id : Z → Z. The Euler class e(ξ ) ∈ H 2 (X; Z) is then defined to be κ ∗ (ι). Since e(ξ ) characterizes the homotopy class of κ, it serves to classify principal S1 bundles over X. A real Euler class e(ξ )Rd is obtained by simply extending coefficients to H 2 (X; Rd ). This is equivalent to composing with the map φ : K(Z, 2) → K(Rd , 2) corresponding to the usual inclusion Z → Rd . Again, we have [X, K(Rd , 2)] ∼ = H 2 (X; Rd ) obtained by pulling back a class ι back to K(Z, 2) gives ι corresponding to id : Rd → Rd . (Note that pulling the class in H 2 (K(Z, 2); Rd ) corresponding to the usual inclusion Z → Rd .) We can now prove the main result using only the standard fibration and pullback theory presented above. Theorem 6.47 The real Euler class e(ξ )Rd is zero if and only if ξ is flat.

6.5

Appendix

Proof With the notation above, consider the following fibration diagram, which we shall use for both parts of the proof. (Here we write BS1 for K(Z, 2) and BSd1 for K(Sd1 , 1) to relate back to the definition of flat bundle.) BSd1

Bi

aC

C

κ

/ BS1 O C

φ

/ K(Rd , 2)

κ

C

X Suppose ξ is flat. Then κ exists by definition and we have φ κ φ Bi κ∗ since φ Bi ∗. But φ κ represents the real Euler class, so we obtain e(ξ )Rd = 0. On the other hand, suppose e(ξ )Rd = 0. Then φ κ ∗. But then the homotopy lifting property of a fibration allows us to factor κ through the fiber as shown in the diagram: Bi κ κ. By definition, this means that the bundle is flat. Of course, when X has finite type, for an integral cohomology class e, we can identify the conditions that e vanishes when coefficients are extended to the reals and that e is torsion. This holds, in particular, when M is a compact manifold. Therefore, we have Corollary 6.48 If X has finite type, then the Euler class e(ξ ) is torsion if and only if ξ is flat. Hence, ξ is flat if and only if e(ξ )Q = 0. Exercises for Chapter 6 Exercise 6.1 There are many kinds of curvatures for manifolds that are derived from the Riemann curvature. The two most important of these are as follows. Definition 6.49 The Ricci curvature is defined to be Ric(X, Y) =

k

R(X, Ei )Y, Ei

i=1

where X and Y are tangent vector fields on Mk and {Ej } is a frame. Recall that a frame {Ej } consists of an orthonormal basis of vector fields in some neighborhood of a point in M. The scalar curvature is defined to be κ=

k j=1

Ric(Ej , Ej ) =

k

R(Ej , Ei )Ej , Ei .

i,j=1

While a manifold Mk is said to have non-negative sectional curvature if sec(X, Y) ≥ 0 for all X and Y, the Ricci curvature is said to be non-negative when Ric(X, X) ≥ 0

269

270

6 : Curvature

for all X. Show that non-negative sectional curvature on M implies non-negative Ricci curvature on M and, in turn, this implies non-negative scalar curvature on M. Hint: First prove the formula Ric(X, X) = X, X

k−1

sec(X, Ej )

j=1

for a frame {Ej } spanning X ⊥ , the space of vector fields orthogonal to X. Exercise 6.2 Let G be a compact (connected) Lie group with bi-invariant metric. It is known that ∇X X = 0 for X any left invariant vector field. Thus, curves whose tangent vectors belong to left invariant vector fields are geodesics. Also, the following identity is known to hold: [X, Y], Z = −Y, [X, Z] for left invariant vector fields X, Y and Z. Here, ·, · is a the bi-invariant metric and [−, −] is the bracket of vector fields (see Exercise A.1). Now, for a compact, connected Lie group G with left invariant vector fields X, Y and Z, show that 1 1. ∇X Y = [X, Y]. (Consider ∇X−Y (X − Y) and use Proposition 6.1 (v).) 2 1 2. R(X, Y)Z = [[X, Y], Z]. 4 (Start with the definition of R, then use (1) to express R completely in terms of brackets. Finally, use the Jacobi identity for the bracket (Exercise A.1 (3)).) 2 1 3. If X and Y are orthonormal, then sec(X, Y) = [X, Y] . 4 Thus, sectional curvature, and hence, Ricci and scalar curvatures are non-negative for a compact, connected Lie group with its natural bi-invariant metric. (Use the definition of sec, (1), Proposition 6.1 (iv) and the identity [X, Y], Z = −Y, [X, Z] with Z = Y.)

The Killing form of a Lie group G is defined to be b(X, Y) = trace(AdX AdY ), where AdZ (W) = [Z, W] is a linear transformation of the vector space of left invariant vector fields. It can be shown that −b is symmetric, bilinear and invariant under all automorphisms of G. If −b is also nondegenerate, in the sense that b(V, W) = 0 for all W implies V = 0, then −b is actually the bi-invariant metric ·, · on G. That is, X, Y = −b(X, Y). A Lie group with nondegenerate Killing form is said to be semisimple and this coincides with Definition 1.41. Using (2) above and the definition of AdZ , show that a semisimple Lie group with metric −b has Ricci curvature 1 Ric(X, Y) = − b(X, Y). 4 Exercise 6.3 Show that (S2n+1 , T, k) is splitting rigid for all T and k.

7

G-spaces

The history of group actions is intimately tied up with the development of both algebraic topology and Lie group theory. In this chapter, we show how minimal models can be used to study compact Lie group actions on manifolds. We focus on certain aspects of group actions that are especially amenable to model techniques. Throughout the chapter, G will generally denote a Lie group acting on a compact manifold M. Certain properties of M will be apparent. For instance, when the action is free (i.e. no point in M is left fixed by any g ∈ G), we have a principal G-bundle M → M/G with the particular consequence that χ (M) = 0. For a general action, we will see that M fits inside the Borel fibration, M → MG → BG, and this will be the central tool in our later study of transformation groups via models. The construction of the Borel fibration will be given in Section 7.2. The efficacy of the general theory is best represented in the case where G is a torus T r = (S1 )r . For G = T r , we essentially focus on issues related to the following two questions: • When does M admit a free or an almost free G-action? • What are the homological and the homotopical properties of the fixed

point set MG of a G-action on M? For instance, here is a conjecture (generally attributed to S. Halperin [129]) that relates the rank of an almost freely acting torus with the dimension of the manifold’s cohomology. Toral rank conjecture (TRC). If M is a nilpotent compact manifold, then dim H ∗ (M; Q) ≥ 2rk0 (M) , where rk0 (M) is the rational toral rank of M. The rational toral rank rk0 (M) is defined as follows. The action of T r on M is said to be almost free if, for any point in the manifold, the subgroup of G fixing that point is a finite group. The largest integer r for which M

272

7 : G-spaces

admits an almost free T r -action is called the toral rank of M, and is denoted rk(M). If M does not admit any almost free torus action, then rk(M) = 0. Unfortunately the invariant rk(M) is not a homotopy invariant and is quite difficult to compute. To obtain a homotopy invariant, we introduce the rational toral rank, rk0 (M); that is, the maximum of rk(Y) among all finite CW complexes Y in the same rational homotopy type as M. An important advantage of using rk0 (M) is that it can be explicitly computed from the minimal model of M. For instance, in the case of a Lie group G, we have rk0 (G) = rank G, and in the case of a homogeneous space G/H, we have rk0 (G/H) = rank G − rank H. In Section 7.3.3, we will prove the toral rank conjecture for certain families of spaces including homogeneous spaces and compact manifolds whose cohomology satisfies the hard Lefschetz property. The TRC can be reformulated as a conjecture on an upper bound for the toral rank: rk0 (M) ≤ log2 dim H ∗ (M; Q) . We will derive other upper bounds for rk0 (M). First, we have rk0 (M) ≤

1 dim H ∗ (M; Q) . 2

Secondly, denoting by Z (L) the center of a graded Lie algebra L, then, when πeven (M) ⊗ Q = 0, we have rk0 (M) ≤ dim Z (π∗ (M) ⊗ Q) . For a nilmanifold, the inequality becomes an equality, and so the rational toral rank of a nilmanifold M with associated Lie algebra L is the dimension of the center of L. The second part of this chapter begins in Section 7.4 and concerns the topology of the fixed point set MG when G is a torus T r . From the Borel fibration M → MG → BG, we see that the algebra H ∗ (MG ; Q) becomes a module over H ∗ (BG; Q). We have seen in Theorem 1.81 that H ∗ (BG; Q) is a polynomial algebra H ∗ (BG; Q) = Q[x1 , . . . , xr ] where, since G = T r , the xi are in degree 2. Let K denote the corresponding field of fractions; that is, the field of rational fractions in the variables x1 , . . . , xr , K = Q(x1 , . . . , xr ). The Borel localization theorem then says that we have an isomorphism of K-vector spaces K ⊗H ∗ (BG;Q) H ∗ (MG ; Q) ∼ = K ⊗ H ∗ (MG ; Q) . We use rational homotopy theory in conjunction with the Borel localization theorem to obtain results concerning the cohomology and the

7.1

Basic definitions and results

homotopy of MG . Concerning cohomology, we prove the equality χ (M) = χ (MG ), and, for each integer n, the inequality dim H p (MG ; Q) ≤ dim H p (M; Q) . p≥n

p≥n

From the first equality we deduce, in particular, that each G-action has a fixed point when χ(M) = 0. To obtain properties of the homotopy groups of the fixed set, we form the Z/2Z-graded vector space V (where (∧V, d) is a Sullivan model for M), with V 0 = πeven (M) ⊗ Q and V1 = πodd (M) ⊗ Q. Then, for each simply connected component F of MG , we prove that there is a differential D on V, D : V 0 → V 1 , and D : V 1 → V 0 such that H 0 (V, D) = πeven (F) ⊗ Q and H 1 (V, D) = πodd (F) ⊗ Q. In particular, if M is rationally elliptic, then each simply connected component F of MG is also rationally elliptic and, moreover, χπ (F) = χπ (M). Beginning in Section 7.6, we consider certain types of actions which are particularly relevant to geometers, the symplectic and Hamiltonian actions on symplectic manifolds. We show how a weakened Lefschetz condition is enough to prove the TRC in the symplectic context. Then we consider the notion of Hamiltonian bundle in both geometric and algebraic formulations and give a proof of a result of Stepien [243] that says Hamiltonian bundles are TNCZ when the fiber is a nilmanifold. Although we are mainly interested in smooth G-manifolds, torus actions can be defined on more general spaces with important results. In particular, this is the case when we consider toral rank and, in Section 7.3, we discuss toral rank in the more general context of finite CW complexes. However, from Section 7.4 onwards, we only consider smooth actions on manifolds. This chapter is essentially an expository chapter. The book of Kawakubo [152] is a good reference for basic topological aspects of group actions. The books of Hsiang [144], Bredon [40] and tom Dieck [252] are the classical references for algebraic topological aspects. Finally, the book of Allday and Puppe [10] is the main reference for the use of rational homotopy theory in the study of transformation groups and the present chapter is, in some sense, preparation for studying that work.

7.1

Basic definitions and results

A topological group G acts on a space X if there is a continuous map ϕ : G × X → X satisfying ex = x and g1 (g2 x) = (g1 g2 )x. Here we have adopted the usual notational convention that gx = ϕ(g, x). When G is a Lie group (which we shall take to be compact unless stated otherwise) and M is

273

274

7 : G-spaces

a manifold, the action is called smooth if ϕ is a smooth map. The manifold M is then called a smooth G-manifold, or for short, simply a G-manifold. The subgroup Gx = {g | gx = x} is called the isotropy subgroup at the point x. The orbit of x is G(x) = {gx| g ∈ G}. Note that the isotropy subgroups corresponding to points in the same orbit G(x) are conjugate because Ggx = gGx g −1 . The canonical map f : G/Gx → G(x) induced by the action on x is clearly a diffeomorphism. We can consider the set of orbits as a topological space by taking the quotient space defined by the following equivalence relation on M: x ∼ y if and only if there exists some g ∈ G with gx = y. This space of orbits is denoted by M/G. A point x ∈ M is a fixed point of the action if Gx = G; that is, gx = x for all g ∈ G. The set of fixed points under the action of G on M is denoted by MG . An action is free if Gx = {e} for all x ∈ M and is almost free if Gx is a finite group for all x ∈ M. Example 7.1 Let G = S1 act by horizontal rotations on the sphere M = S2 . The fixed point set MG is the union of the North and South poles, the orbit space M/G is a semicircle connecting the poles, the isotropy subgroups are G at the fixed poles and the trivial group {e} elsewhere. Note that, in this case, M/G is not a closed manifold, but rather, a manifold with boundary. Basic properties of actions that will be useful for our purposes are summarized in the following proposition. Proposition 7.2 ([152]) If G is a Lie group and M a compact (smooth) G-manifold, then 1. G(x) is a submanifold of M, and the action of G induces a diffeomorphism G/Gx → G(x). 2. Up to conjugacy, there are only finitely many isotropy groups. Therefore there are only finitely many diffeomorphism types of orbits. 3. The fixed point set MG is a possibly disconnected compact submanifold of M whose components may have different dimensions. 4. If the action is free, then M/G is a manifold and the projection M → M/G is a principal G-bundle. The next lemma enables us to reduce the study of fixed sets of torus actions to the study of fixed sets of circle actions. Lemma 7.3 Let T = (S1 )r be a torus acting smoothly on a compact 1 manifold M. Then there is some circle S1 ⊆ T such that MS = MT . Proof A smooth action of a torus T on a compact manifold M has a finite number of orbit types. So let the isotropy groups be G1 , . . . , Gs not including T itself. In the Lie algebra T associated to T, we have the lie algebras of the

7.2

The Borel fibration

isotropy groups: g1 , . . . , gs . Since we don’t include T itself, the gi are all of lower dimension. Therefore there exists a vector in T not in the union of the gi and exponentiation gives a circle S in T which has the property that it only intersects each isotropy group in a discrete (so finite by compactness) subgroup. This is the circle that works. Clearly, MT is in MS . Now assume x is in MS . What is the T-isotropy group Tx at x? It has to be one of the Gi or all of T (in which case it is a fixed point for T). Suppose Tx = Gi . Since x is in MS , we have S contained in Gi , and this contradicts the fact that S intersects Gi in a finite set. Therefore, Tx = T and x is in MT . The same result holds in the context of CW-complexes with finitely many orbit types, see [4, Lemma 4.2.1].

7.2

The Borel fibration

Recall from Section 1.10 that, for every compact Lie group G, there exists a universal principal G-bundle G → EG → BG . The space BG, called the classifying space of G, is EG/G where EG is a contractible space admitting a free G-action. Note that the contractibility of EG combined with the Puppe sequence show that BG G. Example 7.4 When G is the circle S1 , the classifying space BS1 has been constructed in Example 1.74. Recall that the action of the circle on S2n+1 ⊂ Cn+1 is given by the complex multiplication z(z1 , . . . , zn+1 ) = (zz1 , . . . , zzn+1 ). The quotient is the complex projective space CP(n). Let S∞ be constructed as the direct limit of the inclusions S1 ⊂ S2 ⊂ S3 ⊂ . . . ⊂ ∞ ∪∞ i=1 S = S . Now it is known that homotopy groups behave well under direct limits : limj πk (Sj ) = πk (limj (Sj ). Hence πk (S∞ ) = 0 for all k and, indeed, S∞ is contractible. Clearly S∞ is also obtained by taking the direct limit S3 ⊂ S5 . . . ⊂ ∪S2p+1 = S∞ . The free actions of S1 on the odd spheres are compatible with the inclusions, so they induce a free action of S1 on S∞ . Since S∞ is contractible, it is ES1 , and the S1 -bundle S∞ → CP(∞) is the universal S1 -bundle ES1 → BS1 . By multiplying together different copies of the universal bundle S∞ → CP(∞), we obtain the classifying space of an r-torus, BT r = (CP(∞))r . Definition 7.5 Let G be a Lie group and let M be a G-manifold. The associated Borel fibration is the fiber bundle p

M → EG ×G M → BG,

275

276

7 : G-spaces

where EG ×G M is the quotient of EG × M under the action of G defined by g(x, m) = (xg −1 , gm). The projection p associates to [x, m] the class of x in BG; p([x, m]) = [x]. The total space of the Borel fibration, EG ×G M, is usually denoted by MG and its rational cohomology is called the rational def

∗ (M; Q) = H ∗ (M ; Q). equivariant cohomology of M; HG G

Theorem 7.6 1. When the action of G on M is free, then the projection q : MG = EG ×G M → M/G ,

[x, m] → [m] ,

is a homotopy equivalence. 2. When the action of G on M is almost free, then the projection q : MG → M/G is a rational homotopy equivalence. That is, H ∗ (q; Q) : H ∗ (M/G; Q) → H ∗ (MG ; Q) is an isomorphism. 3. When the action of G on M is trivial, then MG = BG × M. 4. When H is a compact subgroup of G, then (G/H)G BH. Proof (1) and (3) are clear while a proof of (2) is outlined in Exercise 7.1. The following sequence of homotopy equivalences (the last one coming from Example 1.79) proves (4): (G/H)G = EG ×G (G/H) ∼ = (EG ×G G)/H ∼ = EG/H BH .

Since (2) above says that, from the viewpoint of rational homotopy theory, almost free actions are equivalent to free actions, it is important to have a criterion to identify them. The following characterization of almost free actions achieves this aim. Theorem 7.7 (Hsiang’s theorem; see [144], [10, Proposition 4.1.7]) Let G be a compact Lie group and let M be a compact G-manifold. Then the group G acts almost freely on M if and only if the rational equivariant ∗ (M; Q), is finite dimensional. cohomology of M, HG Note that, if the action is free, then we have MG M/G, so H ∗ (MG ; Q) ∼ = is finite dimensional because M/G is a compact manifold.

H ∗ (M/G; Q)

7.3

The toral rank

In this section we give conditions for the existence of almost free torus actions on a manifold M. Our first condition concerns the Euler–Poincaré characteristic χ (M).

7.3

The toral rank

Theorem 7.8 If a compact manifold M admits an almost free G-action with G a compact connected Lie group, then the Euler–Poincaré characteristic of M is zero. Of course, for example, any compact Lie group G admits a free action by its maximal torus via translations and we know that χ (G) = 0 since G is rationally a product of odd spheres. On the other hand, by Theorem 7.8, an even dimensional sphere does not admit any almost free circle action. Proof Suppose that G acts almost freely on M. Since BG G, the homotopy fiber of the injection M → EG ×G M has the homotopy type of G. We obtain in this way a homotopy fibration G → M → EG ×G M = MG . The associated Serre spectral sequence satisfies E2 ∼ = H ∗ (MG ; Q) ⊗ H ∗ (G; Q) ⇒ H ∗ (M; Q) . By Theorem 7.7, the term E2 is finite dimensional. Since the Euler characteristic is preserved in a spectral sequence and since χ (G) = 0, we have χ (M) = χ (MG )χ(G) = 0. As we explained in the introduction to this chapter, the main “invariant” for the study of almost free actions is the toral rank of a manifold. Let’s now define this notion in the context of topological spaces. Definition 7.9 The toral rank of a space X, rk(X), is the largest integer r such that a torus T r acts almost freely on X. The toral rank may be bounded for reasons that depend only on topology. For instance, a wedge of spheres has toral rank zero because the basepoint must remain fixed under the action. The latter claim follows simply from the topology of the wedge. The basepoint of the wedge is the only point of the wedge whose removal increases the number of components, so any homeomorphism must fix that point. Unfortunately, the toral rank is not a homotopy invariant. Example 7.10 (Free S1 -action on Y S2 ∨ S3 ∨ S3 ) Let S1 act freely on S3 by the Hopf action, and on S1 × S2 by translation on the first factor. In each space select a free S1 -orbit and glue together S3 and S1 × S2 along these orbits. The result is a three dimensional CW complex Y admitting a free circle action. Since the orbits in S3 can be contracted to a point, we see that Y S3 ∨ ((S1 × S2 )/(S1 × ∗)). The second summand, which for convenience we denote by Z, is the half-smash product of S1 and S2 . We

277

278

7 : G-spaces

also have (from a homotopy pushout argument) Z = (S1 × S2 )/(S1 × ∗) (S1 × S1 )/(S1 × ∗). But the half-smash product (S1 × S1 )/(S1 × ∗) is easy to describe; it is simply a torus with a meridian collapsed and, homotopically, this is a 2-sphere with North and South poles identified or, equivalently, a 2-sphere with an arc attached to the North and South poles. This last identification is clearly of the homotopy type of S2 ∨ S1 , so we see that Z = (S1 × S1 )/(S1 × ∗) (S2 ∨ S1 ) = S3 ∨ S2 . Hence, the CW complex Y has the homotopy type of the wedge S2 ∨S3 ∨S3 . By the discussion above, rk(S2 ∨ S3 ∨ S3 ) = 0, but rk(Y) ≥ 1. Remark 7.11 We can also define the integer rks (M) to be the maximal r such that there is a torus T r that acts smoothly on M. This invariant obeys rks (M) ≤ rk(M), but it is very difficult to compute and the exact relationship does not seem to be known. The “almost free” condition is not a standard condition in transformation groups, but work of R. Schultz [233] has produced, for instance, an exotic sphere N of dimension 17 that admits no free smooth S1 -action. Of course, N does admit a continuous free circle action since N and S7 are homeomorphic. For general remarks concerning the difference between smooth and continuous degrees of symmetry, see [144, Chapter VII]. Since we are interested in obtaining a homotopy invariant and since we are only considering the rational homotopy type of spaces, we are led to the following variation of the definition of the toral rank. Definition 7.12 The rational toral rank of a space X, rk0 (X), is the maximum of rk(Y) for all finite CW complexes Y in the rational homotopy type of X.

7.3.1

Toral rank for rationally elliptic spaces

The first computations on the rational toral rank are due to Allday and Halperin [6]. Recall that, if M is a rationally elliptic space, the homotopy Euler characteristic is defined by χπ (M) = rank πeven (M) − rank πodd (M). The first toral rank estimate derived from algebraic models was also one of the first verifications (along with formality for Kähler manifolds and the solution of the closed geodesic problem) that rational homotopy theory could be used to obtain interesting geometric information.

7.3

The toral rank

Theorem 7.13 ([6]) If M is a nilpotent rationally elliptic space, then rk0 (M) ≤ −χπ (M) . Proof Consider the homotopy fibration T r → M → MT r associated to an almost free action of T r on M. Since M and T r are rationally elliptic spaces, the same is also true for MT r . This follows for homotopy from the long exact homotopy sequence associated to a fibration and, for cohomology, from Theorem 7.7. The long exact homotopy sequence then gives χπ (M) = χπ (MT r ) + χπ (T r ). Recall now that, for a rationally elliptic space X, we have χπ (X) ≤ 0 (see Theorem 2.75). From χπ (MT r ) ≤ 0 and χπ (T r ) = −r, we deduce r ≤ −χπ (M). Corollary 7.14 If G is a compact connected Lie group, then rk0 (G) = rank G. Proof Recall from Example 2.39 that the minimal model of a Lie group G is an exterior algebra (∧(y1 , . . . , yr ), d = 0) with r = rank G. It follows that −χπ (G) = rank G. Therefore rk0 (G) ≤ rank G. On the other hand, if T r is a maximal torus in G, then the left multiplication by T r yields a free action of T r on G, so that rk0 (G) ≥ rank G. Corollary 7.15 If M = G/K is the quotient of a compact connected Lie group G by a compact connected subgroup K, then rk0 (G/K) = rank G − rank K . Proof From the fibration K → G → G/K (and its long exact homotopy sequence), we deduce χπ (G/K) = χπ (G) − χπ (K) = −rank G + rank K. This implies the inequality rk0 (G/K) ≤ −χπ (G/K) = rank G − rank K . Now denote by T s a maximal torus in G and by T r a maximal torus in K. The maximal tori of a compact Lie group are all conjugate, so we can suppose T r ⊂ T s . The left multiplication by T s−r = T s /T r on G/K is again a free action. Therefore the rational toral rank of G/K is greater than or equal to the difference s − r, rk0 (G/K) ≥ rank G − rank K. Example 7.16 As we have seen, an odd dimensional sphere S2n−1 admits a free circle action. Since χπ (S2n−1 ) = −1, we have rk0 (S2n−1 ) = 1.

279

280

7 : G-spaces

7.3.2

Computation of rk 0 (M) with minimal models

Let G = T r be an r-torus, M a nilpotent compact G-manifold, and p

M → EG ×G M → BG the associated Borel fibration. Our first aim in this section is the description of the relative minimal model of the Borel fibration. Let’s first recall the rational cohomology of T r and BT r (Theorem 1.81). The cohomology of T r is an exterior algebra on r generators in degree 1, H ∗ (T r ; Q) = ∧(y1 , . . . , yr ) , and the cohomology of the classifying space BT r is a polynomial algebra on r generators in degree 2, H ∗ (BT r ; Q) = Q[x1 , . . . , xr ] . In particular, the minimal model of T r is (∧(y1 , . . . , yr ), 0), and the minimal model of BT r is (∧(x1 , . . . xr ), 0). Let (∧V, d) be the minimal model for M. Then a relative minimal model for the Borel fibration is (∧(x1 , . . . , xr ), 0) → (∧(x1 , . . . xr ) ⊗ ∧V, D) → (∧V, d) . When the action is almost free, H ∗ (∧(x1 , . . . , xr ) ⊗ ∧V, D) is finite dimensional by Theorem 7.7. This yields a characterization of almost free actions in terms of minimal models. Proposition 7.17 ([130, Proposition 4.2]) If M is a nilpotent compact mdimensional manifold, then rk0 (M) ≥ r if and only if there is a relative minimal model of the form (∧(x1 , . . . , xr ), 0) → (∧(x1 , . . . , xr ) ⊗ ∧V, D) → (∧V, d) , where |xi | = 2 for i = 1, . . . , r, (∧V, d) is the minimal model of M and the cohomology H ∗ (∧(x1 , . . . , xr ) ⊗ ∧V, D) is finite dimensional. Moreover, if rk0 (M) ≥ r, then T r acts freely on a finite CW complex X that has the same rational homotopy type as M, and if m − r ≡ 0 mod 4, then we can choose X to be a compact manifold. Proof If we have an almost free T r action on M, the discussion preceding the statement of the proposition shows that we have a relative minimal model with the required properties. Conversely, suppose we have such a relative minimal model, and let N be a finite nilpotent CW complex whose minimal model is (∧(x1 , . . . , xr ) ⊗

7.3

The toral rank

∧V, D) (see Section 2.6.1). The classes [xi ] define classes in H 2 (N; Q). Multiply by integers ki to obtain [ki xi ] ∈ H 2 (N; Z) and use these classes to define a map ϕ : N → K(Z, 2)r = (CP(∞))r . The pullback of the universal T r -bundle with basis (CP(∞))r gives a finite nilpotent CW complex X in the rational homotopy type of M that admits a free T r -action. The cohomology of (∧(x1 , . . . , xr )⊗∧V, D) satisfies Poincaré duality and the fundamental class is in degree m − r. Therefore by Theorem 3.2, when m − r ≡ 0 mod 4 we can choose N to be a compact manifold. In this case, the classifying map may be compressed into CP(m − r) and then may be replaced by a smooth map. The pullback of the Hopf principal bundle over CP(m − r) then is a principal circle bundle whose total space is a smooth manifold of the rational homotopy type of M. Example 7.18 Denote by S11 → N → S12 the sphere bundle associated to the tangent bundle of the sphere S12 . We then denote by S11 → E → (S3 )4 the pullback of this fibration along a map (S3 )4 → S12 of degree one. Let’s show that rk0 (E) = 1. First of all, the minimal model of E is (∧(y1 , y2 , y3 , y4 , z), d) , |yi | = 3 , |z| = 11 , dz = y1 y2 y3 y4 . We then consider the relative minimal model (with |x| = 2), (∧x, 0) → (∧x ⊗ ∧(y1 , y2 , y3 , y4 , z), D) → (∧(y1 , y2 , y3 , y4 , z), d) where D(z) = dz + x6 . Clearly, the cohomology H ∗ (∧x ⊗ ∧(y1 , y2 , y3 , y4 , z), D) is finite dimensional, which implies by Proposition 7.17 that there exists a manifold in the rational homotopy type of E that admits a free action of S1 . In particular rk0 (E) ≥ 1. Now suppose that rk0 (E) = r ≥ 1. By Proposition 7.17, there exists a relative minimal model (∧(x1 , . . . , xr ), 0) → (∧(x1 , . . . , xr , y1 , y2 , y3 , y4 , z), D) → (∧(y1 , y2 , y3 , y4 , z), d) , with dim H ∗ (∧(x1 , . . . , xr , y1 , y2 , y3 , y4 , z), D) < ∞. Suppose D(y1 ) = a = 0, a ∈ ∧(x1 , . . . , xr ). Now, D(z) = dz + τ = y1 y2 y3 y4 + τ with τ ∈ Ideal(xi ), so 0 = D2 (z) = ay2 y3 y4 + ξ , where ξ belongs to the ideal generated by y1 and ∧≥3 (x1 , . . . xr ) for degree reasons. This is impossible because ay2 y3 y4 is not in this ideal. Therefore, by similar arguments, D(y1 ) = D(y2 ) = D(y3 ) = D(y4 ) = 0.

281

282

7 : G-spaces

Now write D(z) = α + β with α ∈ ∧(x1 , . . . , xr ) and β ∈ ∧(xi ) ⊗ ∧+ (yj ). The projection (∧(x1 , . . . , xr , y1 , y2 , y3 , y4 , z), D) → (∧(x1 , . . . , xr )/(α), 0) is then surjective in cohomology. Therefore ∧(x1 , . . . , xr )/(α) is finite dimensional and this can only happen when r ≤ 1. Finally, note that, by Theorem 7.13, we have an upper bound for rk0 (E) given by 5 = −χπ (E). Therefore, we see that the Allday–Halperin estimate is definitely not sharp. Example 7.19 In [148], an example, due to Halperin, is given of manifolds M and N with rk0 (M) = rk0 (N) = 0 and rk0 (M × N) ≥ 1. This shows, at least rationally, that taking products can increase the symmetry of manifolds. The minimal models of M = S12 and N are, respectively, (∧(x, y), d), |x| = 12, |y| = 23, dy = x2 and (∧(u1 , u2 , u3 , u4 , u5 , u6 , v, w), d) with |u1 | = |u2 | = |u3 | = |u4 | = 3, |u5 | = 5, |u6 | = 19, |w| = 18, |v| = 35, dv = w2 + u1 u2 u3 u4 u5 u6 , with the other differentials being zero. Since rk0 (M) ≤ −χπ (M) = 0, we have rk0 (M) = 0. We will show that the cohomology of any Sullivan algebra of the form (∧a ⊗ ∧(u1 , u2 , u3 , u4 , u5 , u6 , v, w), D) is infinite dimensional when |a| = 2 and the image of D − d belongs to the ideal generated by a. This will imply that rk0 (N) = 0. For degree reasons D must satisfy the following properties Du1 = α1 a2 , Du2 = α2 a2 , Du3 = α3 a2 , Du4 = α4 a2 , Du5 = α5 a3 Du6 = α6 a10 + µaw + a3 P Dv = w2 + u1 u2 u3 u4 u5 u6 + waF + u6 a3 R + T Dw ∈ ∧+ (a) ⊗ ∧+ (u1 , u2 , u3 , u4 , u5 ) , with α1 , α2 , α3 , α4 , α5 , α6 , µ ∈ Q, R and P ∈ ∧a ⊗ ∧+ (u1 , u2 , u3 , u4 , u5 ), and F and T in ∧+ (a) ⊗ ∧(u1 , u2 , u3 , u4 , u5 ). Looking at the coefficient of u6 in D2 v, we find 5 i · · · u6 − u6 a3 D(R) = 0, (−1)i+1 αi a2 u1 · · · u i=1

which implies that α1 = · · · = α5 = 0. The equation 0 = D2 u6 gives µaDw = 0, so that either µ = 0 or else Dw = 0. Now let’s consider the equation D2 v = 0. We have 0 = 2wDw − u1 u2 u3 u4 u5 (α6 a10 + µaw + a3 P) + D(w)aF + (α6 a10 + µaw + a3 P)a3 R . If µ = 0, then the coefficient of w becomes Dw, so that Dw = 0. If Dw = 0. In conclusion, in 0, the coefficient of u1 u2 u3 u4 u5 aw is µ so that µ = i any case we have µ = 0 and Dw = 0. Write R = m i=0 a Ri with Ri ∈

7.3

The toral rank

∧+ (u1 , · · · , u5 ), and observe that the coefficient of a13 in D2 v is α6 R. It follows that α6 = 0. Because of the calculations above, the projection q : (∧(a, u1 , u2 , u3 , u4 , u5 , u6 , v, w), D) → (∧(a, w)/(w2 ), 0) obtained by mapping the ui and v to 0 is a cdga morphism. If the cohomology H ∗ (∧(a, u1 , u2 , u3 , u4 , u5 , u6 , v, w), D) is finite dimensional, then some power ak is a coboundary. But this means that, in cohomology, 0 = [a]k → ak = 0 which is impossible. Therefore rk0 (N) = 0. We now show that rk0 (M ×N) = 1 by giving an explicit relative minimal model that has finite dimensional cohomology: (∧(a, x, y, u1 , u2 , u3 , u4 , u5 , u6 , v, w), D) . The differential is defined as follows Du1 = Du2 = Du4 = Du3 = Dw = 0, Dx = u1 u2 u3 a2 , Dy = x2 + 2u1 u6 a, Du5 = a3 , Du6 = u2 u3 xa, Dv = w2 + u1 u2 u3 u4 u5 u6 + xau4 u6 + ya2 u2 u3 u4 . A straightforward computation shows that D2 = 0. To prove that the cohomology is finite dimensional, we put a new gradation V (p) on V by letting |u1 | = |u2 | = |u3 | = |u4 | = |u6 | = 3, |a| = |x| = |w| = 2, |y| = |v| = 3 and |u5 | = 5. Since D(V (p) ) ⊂ (∧V)≥p+1 , this gives a spectral sequence converging to the cohomology of the algebra. The E2 term is isomorphic to ∧(u1 , u2 , u3 , u4 , u6 ) ⊗ (∧x)/(x2 ) ⊗ (∧a)/(a3 ) ⊗ (∧w)/(w2 ) and is finite dimensional. This proves that the cohomology is finite and that rk0 (M × N) ≥ 1.

7.3.3

The toral rank conjecture

The main problem related to the toral rank is the so called toral rank conjecture (TRC), which is usually attributed to S. Halperin. Conjecture 7.20 (TRC conjecture) Let M be a nilpotent finite CW complex. Then dim H ∗ (M; Q) ≥ 2rk0 (M) . The intuition behind the TRC is that an almost free T r -action somehow injects T r , at least cohomologically, into M. Of course, dim H ∗ (T r ; Q) = 2r , so the conjecture is an expression of this intuitive notion. To set the record

283

284

7 : G-spaces

straight, however, we note that examples exist showing that H ∗ (M; Q) does not necessarily contain an exterior algebra on r generators even when an almost free T r -action exists on M (see [4, Remarks 4.4.2 (5)]). From Proposition 7.17 we obtain a reformulation of the TRC conjecture in terms of minimal models. Conjecture 7.21 (Algebraic TRC) Let (∧V, d) be a minimal cdga, and let (∧(x1 , . . . , xr ), 0) → (∧(x1 , . . . , xr ) ⊗ ∧V, D) → (∧V, d) be a relative minimal model with |xi | = 2, and such that the cohomology algebra H ∗ (∧(x1 , . . . , xr )⊗∧V, D) is finite dimensional. Then dim H ∗ (∧V, d) ≥ 2r . The TRC conjecture is open in general, but has been proved in some interesting cases that we will now consider. Proposition 7.22 The TRC is true for any product of odd-dimensional spheres. Proof Suppose an r-torus T r acts almost freely on a product of odddimensional spheres M = Sn1 × Sn2 × . . . × Snp . By Theorem 7.13, r ≤ −χπ (M) = p . This implies that 2r ≤ 2p = dim H ∗ (M; Q) .

Proposition 7.23 The TRC is true for homogeneous spaces. Proof Let G be a compact connected Lie group and let K ⊂ G be a compact connected subgroup. Denote the rational toral rank of G/K by r, and consider the Serre spectral sequence associated to the fibration K → G → G/K: H ∗ (G/K) ⊗ H ∗ (K) ⇒ H ∗ (G). It follows that dim H ∗ (G) ≤ dim H ∗ (G/K) · dim H ∗ (K). Since r = rank G − rank K (see Corollary 7.15), this gives 2r = 2rank G−rank K =

2rank G dim H ∗ (G) ≤ dim H ∗ (G/K) . = dim H ∗ (K) 2rank K

Recall that a connected graded commutative algebra H satisfies the hard Lefschetz property if H behaves like the cohomology of a compact Kähler

7.3

The toral rank

manifold (see Theorem 4.35). This means that the following property is satisfied: • There is an element ω ∈ H 2 such that for every p < m, the multiplication by ωm−p induces an isomorphism H p → H 2m−p . Theorem 7.24 ([8]) If a torus T r acts almost freely on a compact connected manifold M with a cohomology satisfying the hard Lefschetz property, then the injection of an orbit, T r · x → M, induces an injection in homology. In particular, the TRC is true in this case. Proof We consider the Borel fibration associated to the action of T r on M: M → MT r → BT r . The connecting map of the fibration, δ : T r ∼ = BT r → M corresponds to the injection of an orbit of the action. We prove that H∗ (δ; Q) is an injective map. Write H ∗ (BT r ; Q) = Q[x1 , . . . , xr ] and note that the image of the transgression H 1 (M; Q) → H 2 (BT r ; Q) is a sub-vector space of the vector space generated by the xi . By making a change of generators, we suppose that the image is the vector space generated by the elements xs+1 , xs+2 , . . . , xr . The quotient map q : (∧(x1 , . . . , xr ), 0) → (∧(x1 , . . . , xs ), 0) defined by xj for j ≤ s q(xj ) = 0 otherwise can be realized by a map g : BT s → BT r . We now pull back the Borel fibration along g to obtain a fibration with base BT s . MO =

M

/ MT r O f

/ E

/ BT r O g

/ BT s

By construction, the transgression of the new fibration, H 1 (M; Q) → H 2 (BT s ; Q), is zero. This implies by Corollary 4.41 that the Serre spectral sequence of the fibration M → E → BT s collapses at the E2 term: H ∗ (E; Q) ∼ = H ∗ (M; Q) ⊗ H ∗ (BT s ; Q). In particular if s > 0, then H ∗ (E; Q) is infinite dimensional. Notice now that the homotopy fiber of the induced map f : E → MT r is the homotopy fiber of g, T r−s . Since H ∗ (T r−s ; Q) and H ∗ (MT r ; Q) are finite dimensional (the latter by Theorem 7.7), the same is true for H ∗ (E; Q) and, thus, s = 0. This means that in the original Borel fibration, the transgression H 1 (M; Q) → H 2 (BT r ; Q) is a surjective map. Now let (∧(x1 , . . . , xr ), 0) → (∧(x1 , . . . , xr ) ⊗ ∧W, D) → (∧W, d)

285

286

7 : G-spaces

be a relative minimal model for the Borel fibration. By what we have said above about the transgression, we obtain a decomposition ∧W = ∧(z1 , . . . , zr ) ⊗ ∧V, with D(zi ) = xi . We now let x¯ 1 , . . . , x¯ r be variables in degree 1, and we form the cochain algebra (C, D) = (∧(x1 , . . . , xr , x¯ 1 , . . . , x¯ r ), D) where D(xi ) = 0 and D(x¯ i ) = xi . This cochain algebra is contractible, which means that H ∗ (C, D) = Q. The projection π : (C, D) ⊗(∧(x1 ,...,xr )) (∧(x1 , . . . , xr ) ⊗ ∧W, D) → (∧W, d) , obtained by mapping the xi and the x¯ i to zero, is a morphism of differential graded algebras. Since (C, D) is contractible, π is a surjective quasi-isomorphism, and since (∧W, d) is minimal, π admits a section σ . The section is not unique, but in any case we always have σ (zi ) = zi − x¯ i . We denote by ψ : (∧(z1 , · · · , zr ), 0) → (∧W, d) the canonical injection, and we define a morphism of commutative differential graded algebras ϕ : (C, D) ⊗(∧(x1 ,...,xr )) (∧(x1 , . . . , xr ) ⊗ ∧W, D) → (∧(x¯ 1 , · · · , x¯ r ), 0) by sending the xi , the zi and V to 0, and x¯ i to x¯ i . The composition ϕ ◦ σ ◦ ψ maps zi to −x¯ i , and is therefore an isomorphism. This shows that H ∗ (ϕ) : H ∗ (M; Q) → ∧(x¯ 1 , . . . x¯ n ) is a surjective map. We now observe that the relative minimal model ϕ

(∧(xi ) ⊗ ∧W, D) → (C, D) ⊗(∧(xi )) (∧(xi ) ⊗ ∧W, D) −→ (∧(x¯ i ), 0) is a model for the homotopy fibration T r ∼ = BT r → M → MT r . This implies that ϕ is a model for the injection of an orbit, and so the injection of an orbit induces as well an injection in rational homology. Corollary 7.25 A compact simply-connected manifold M whose cohomology satisfies the hard Lefschetz property does not admit any almost free torus action. Proof If there were an almost free T r -action on M, then there would exist an injection H∗ (T r ; Q) → H∗ (M; Q). Since M is simply connected this is impossible. Note that, in the case of a symplectic manifold, this follows from Subsection 7.6.2 (in particular Theorem 7.67) as well. A weaker form of the TRC has been proved by Allday and Puppe in [11]. Theorem 7.26 If a torus T r acts almost freely on a compact nilpotent manifold M, then dim H ∗ (M; Q) ≥ 2r .

7.3

The toral rank

Proof Let us consider the relative minimal model of the Borel fibration: (∧(x1 , · · · , xr ), 0) → (∧(x1 , · · · , xr ) ⊗ ∧V, D) → (∧V, d) . We decompose ∧V into the direct sum ∧V = B ⊕ H ⊕ R, with d(H ⊕ B) = 0 and d : R → B an isomorphism. The graded vector space E = (∧(x1 , · · · , xr ) ⊗ R) ⊕ D(∧(x1 , · · · , xr ) ⊗ R) is then a free acyclic ∧(x1 , · · · , xr )-module. The quotient map is therefore a quasi-isomorphism of ∧(x1 , . . . , xr )-modules,

(∧(x1 , · · · , xr ) ⊗ ∧V, D) −→ (∧(x1 , · · · , xr ) ⊗ ∧V, D)/E ∼ = (∧(x1 , · · · , xr ) ⊗ H, D) . For every element h of H we write D(h) = α(h) + β(h) ,

α(h) ∈ ∧(x1 , · · · , xr ), β(h) ∈ ∧(x1 , · · · , xr ) ⊗ H + .

Because of the finiteness of H ∗ (∧(x1 , . . . , xr ) ⊗ H, D), there exists at least r linearly independent elements hi with α(hi ) = 0. The degrees of these elements hi are odd. By construction, H is isomorphic to H ∗ (∧V, d) = H ∗ (M; Q). Since the Euler–Poincaré characteristic of M is zero by Theorem 7.8, the dimension of the rational cohomology of M is at least 2r.

7.3.4

Toral rank and center of π∗ (M) ⊗ Q

There is a strong relation between the rational toral rank of a space and the dimension of the center, Z (LM ) of the rational homotopy Lie algebra LM = π∗ (M) ⊗ Q. Proposition 7.27 ([7]) If M is a nilpotent compact manifold with πeven (M)⊗ Q = 0, then rk0 (M) ≤ dim Z (LM ) . Proof We suppose we have an almost free action of T r on M and we denote by (∧(x1 , · · · , xr ), 0) → (∧(x1 , · · · , xr ) ⊗ ∧V, D) → (∧V, d) the relative minimal model of the associated Borel fibration M→MT r →BT r . By hypothesis, V = V odd . Therefore D(V) ⊂ ∧(x1 , · · · , xr ) ⊗ ∧even (V). For v ∈ V, we write D(v) = α(v) + β(v), with α(v) ∈ ∧(x1 , · · · , xr ) , β(v) ∈ ∧(x1 , · · · , xr ) ⊗ ∧+ V .

287

288

7 : G-spaces

By sending V to zero, we obtain a map of differential graded algebras (∧(x1 , · · · , xr ) ⊗ ∧V, D) → (∧(x1 , · · · , xr )/(α(V)), 0) . Since the xi are cocycles, this map is surjective in cohomology. Now, because the action is almost free, the algebra H ∗ (∧(x1 , · · · , xr ) ⊗ ∧V, D) is finite dimensional, so dim (∧(x1 , · · · , xr )/α(V)) < ∞. This implies that the ideal α(V) has at least s independent generators α(v1 ), . . . , α(vs ) with s ≥ r. We denote by W the graded sub-vector space elements of V consisting of the + v such that α(v) belongs to the ideal i α(vi ) · ∧(x1 , . . . , xr ) . Clearly, V = W ⊕ (v1 , . . . , vs ). We now prove that d1 (V) ⊂ ∧2 (W), where d1 denotes the quadratic part of the differential d. This implies, by Proposition 2.60, that dim Z (LM ) ≥ s ≥ r. Let’s fix some i ≤ s. For v ∈ V, d1 (v) can be decomposed into a sum, d1 (v) = vi σ + ω, where σ and ω belong to the algebra generated by W ⊕ (v1 , . . . , vî , . . . , vs ). Then a standard, but tedious, computation shows that D2 (v) − α(vi ) ⊗ σ belongs to ⎛ ⎞

α(vj )R ⊗ ∧V ⎠ , R ⊗ ∧≥2 V ⊕ (α(vi )R+ ⊗ ∧V) ⊕ ⎝ i =j

with R = ∧(x1 , . . . , xr ). Since D2 (v) = 0, we have σ = 0 and d1 (V) ⊂ ∧2 (W). For nilmanifolds the relation is stronger. Recall from Theorem 3.18 that a nilmanifold M with associated Lie algebra L has for minimal model the cochain algebra on L, C ∗ (L), and that L is also the rational homotopy Lie algebra of M. Denote Z (L) the center of L. Then we have: Theorem 7.28 Let M be a nilmanifold with associated Lie algebra L. Then rk0 (M) = dim Z (L). Proof By Proposition 7.27, rk0 (M) ≤ dim Z (LM ) = dim Z (L). We now show that there exists a principal T s -bundle, T s → M → N, with s = dim Z (L). This provides a free T s -action on M and shows that rk0 (M) ≥ dim Z (L). Indeed the short exact sequence 0 → Z (L) → L → L/Z (L) → 0 induces a relative minimal model C ∗ (L/Z (L)) → C ∗ (L) → C ∗ (Z (L)) . By definition of the cochain algebra on a Lie algebra, C ∗ (Z (L)) = (∧(x1 , . . . xs ), 0) and C ∗ (L) = (∧(y1 , . . . , ym , x1 , . . . , xs ), D) with D(xi ) =

7.3

The toral rank

Pi ∈ ∧2 (y1 , . . . , ym ). Let u1 , . . . , us be elements of degree 2. We then have a natural morphism of differential graded algebras ϕ : (∧(u1 , . . . , us ), 0) → (∧(y1 , . . . , ym ), d) given by ϕ(us ) = Ps (where the Ps are d-cocycles because D = d on ∧(y1 , . . . , ym )). The morphism ϕ is realized by a map f : N → BT s where N is the nilmanifold associated to the Lie algebra L/Z (L). A computation with minimal models shows that the nilmanifold M has the rational homotopy type of the pullback along f of the universal T s -bundle: / ET s

M N

f

/ BT s

7.3.5

The TRC for Lie algebras

By Theorem 7.28, for finite dimensional nilpotent Lie algebras, the TRC conjecture reduces to a conjecture on the center of the Lie algebra. Conjecture 7.29 (Lie algebra TRC) If L is finite dimensional nilpotent Lie algebra defined over the rational numbers, then dim H ∗ (L; Q) ≥ 2dim Z (L) . While the conjecture is open in general, it has been proved for nilpotent Lie algebras with certain restricted structures. For instance, we have the following result of Deninger and Singhof (see [73]; also [48]). Theorem 7.30 The Lie Algebra TRC conjecture is true for 2-step nilpotent Lie algebras. Proof A Lie algebra L is called 2-step if there is a short exact sequence 0 → Z (L) → L → L/Z (L) → 0, where Z (L) is the center of L and L/Z (L) is abelian. The cochain algebra on the Lie algebra L therefore has the form C ∗ (L) = (∧U ⊗ ∧V, d) with d(U) = 0, and d(V) ⊂ ∧2 U. By construction, (∧V, 0) is the cochain algebra on Z (L). We also have U = U 1 and V = V 1 . We equip C ∗ (L) =

289

290

7 : G-spaces

(∧U ⊗ ∧V, d) with a second (lower) gradation by putting U = U0 and V = V1 . Then the differential d satisfies d(C ∗ (L))p ⊂ (C ∗ (L))p−1 , q

q+1

q

and the cohomology is bigraded. It follows that ⊕p+q=r C ∗ (L)p is a subcomplex and, by the usual property of the Euler–Poincaré characteristic of a finite complex, we have the equality of polynomials ⎞ ⎛ ⎞ ⎛ q q ⎝ ⎝ (−1)p dim C ∗ (L)p ⎠ t r = (−1)p dim Hp (C ∗ (L))⎠ t r . r≥0

p+q=r

r≥0

p+q=r

Denote this polynomial by PL (t). Note that q is the usual degree, so the summation index r = q + p mixes together the degree and lower degree. Since C ∗ (L) = ∧U ⊗ ∧V, we have PL (t) = (1 + t)m (1 − t 2 )n ,

m = dim U,

n = dim V .

Now multiply the last equation by (1 − t)m to obtain (1 − t)m PL (t) = (1 − t 2 )m+n . By taking the value at the complex number i, we see that |PL (i)| ≥ 2n . Since the modulus of PL (t) at the point i is less than or equal to the dimension of H ∗ (C ∗ (L)) (by the triangle inequality), we obtain dim H ∗ (C ∗ (L)) ≥ |PL (i)| ≥ 2n = 2Z (L) .

Of course, we immediately obtain the corresponding result for nilmanifolds with appropriately restricted fundamental group. Corollary 7.31 The TRC conjecture is true for nilmanifolds M with 2-nilpotent fundamental group. The TRC conjecture is also true for low dimensional nilpotent Lie algebras. In [48], it is proved that the TRC conjecture is true for L if one of the following conditions holds: 1. dim L ≤ 14; 2. dim Z (L) ≤ 5; 3. dim L/Z (L) ≤ 7.

7.4

7.4

The localization theorem


There is a long history in topology and geometry of obtaining global information from local properties. From the Lefschetz theorem to the Poincaré–Hopf theorem to the recent theory of stationary phase, the dynamical philosophy has been that the “interesting points” of a geometricdynamical system are the stationary points. Compact group actions on manifolds fit into this philosophical framework through their accompanying fundamental vector fields, so we should expect interesting relations between the algebraic topological invariants of the global manifolds and the fixed sets of the actions. The aim of this section is to relate properties of the cohomology of MG to properties of the cohomology of M when G is a compact Lie group acting on a compact manifold M. Recall first that if A is a commutative ring without zero divisors, then its field of fractions K is the quotient of A × (A\{0}) by the equivalence relation (a, b) ∼ (c, d) if ad = bc . For instance, the field of fractions of Z is Q, and the field of fractions of the polynomial algebra Q[x1 , . . . , xr ] is the field of rational functions in the variables x1 , . . . , xr , denoted by K = Q(x1 , . . . , xr ) . If P is an A-module (or an A-algebra), we can consider the K-vector space (or the K-algebra) K ⊗A P. The kernel of the localization map ϕ : P → K ⊗A P ,

ϕ(p) = 1 ⊗ p ,

is the submodule consisting of elements p such that, for some a ∈ A, we have ap = 0. We denote by A = Q[x1 , . . . , xr ] = H ∗ (BG; Q) the rational cohomology of BG and by K its field of fractions. Note that, since the elements xi have even degree, the tensor product K ⊗A V inherits only a Z/2Z gradation (i.e. odd or even) for every graded A-module V. In other words, the even elements of V remain in even degree and the odd degree elements remain in odd degree.

291

292

7 : G-spaces

Let’s now come back to our G-manifold M. The injection of F = MG into M induces a morphism f between Borel fibrations / M

F BG × F = EG ×G F BG

f

=

/ EG ×G M / BG

We are now ready to recall Borel’s localization theorem. Keep in mind that A = Q[x1 , . . . , xr ] = H ∗ (BG; Q) and K denotes the associated field of fractions. Theorem 7.32 ([144, Proposition 1, page 45]; also [10]) If G is a torus and M is a compact G-manifold, then the map induced from f by localization, ∗ ∗ K ⊗A H ∗ (f ) : K ⊗A HG (M; Q) → K ⊗A HG (F; Q) = K ⊗ H ∗ (F; Q) ,

is an isomorphism of K-algebras.

7.4.1

Relations between G-manifold and fixed set

As a corollary to Theorem 7.32, we deduce the fundamental relation between the Euler–Poincaré characteristics of M and of MG that was already mentioned in Proposition 3.32. Theorem 7.33 Let G be a torus acting on a compact manifold M. Then χ(MG ) = χ(M) . Proof Write A = H ∗ (BG; Q) and K its field of fractions. The Serre spectral sequence associated to the fibration M → EG ×G M → BG satisfies p,q p+q E2 ∼ = Ap ⊗ H q (M; Q) =⇒ HG (M; Q) .

We tensor the terms of the spectral sequence by K over A and obtain a spectral sequence of finite dimensional K-vector spaces. This gives the following sequence of equalities: ∗ (M)) . χ (M) = χK (K ⊗A E2 ) = χK (K ⊗A E∞ ) = χK (K ⊗A HG

By Borel’s localization theorem, the last term is equal to χK (K⊗H ∗ (MG )) = χ (MG ) .

7.4


Remark 7.34 Borel’s localization theorem is not true for any Lie group G. For instance consider the action of SO(3) on S3 ⊂ R4 given by rotations in the three first coordinates. The fixed point set is composed of the two poles (0, 0, 0, 1) and (0, 0, 0, −1). Therefore χ (S3 ) = 0 and χ (S3 )G = 2 in contradiction to a possible generalization of Theorem 7.33. A very useful criterion for the existence of fixed points may be obtained from the localization theorem. Theorem 7.35 (Borel fixed point criterion) Suppose M is a compact Gmanifold where G is a torus. Then the action has fixed points if and only if the homomorphism H ∗ (BG; Q) → H ∗ (MG ; Q) is injective. Proof Write A = H ∗ (BG; Q) and let K be its field of fractions. If H ∗ (BG; Q) → H ∗ (MG ; Q) is injective, then the A-module generated by 1 ∈ H 0 (MG ; Q) is free, and the K-module K ⊗A H ∗ (MG ; Q) is nonzero. Therefore K ⊗ H ∗ (MG ; Q) is also nonzero, so MG cannot be empty. On the other hand, if MG is not empty, the choice of a point in MG gives a map BG → BG × MG that is a section for the composition BG × MG = f

p

EG ×G MG → EG ×G M → BG. In particular, H ∗ (BG; Q) → H ∗ (MG ; Q) admits a retraction and is therefore injective. i

Recall from Definition 4.39 that a fibration F → E → B is TNCZ (totally noncohomologous to zero) with coefficients in L if i∗ : H ∗ (E; L) → H ∗ (F; L) is surjective. Theorem 7.36 Suppose M is a compact G-manifold, where G is a torus. Then the Borel fibration, M → MG → BG, is TNCZ with coefficients in Q if and only if dim H ∗ (M; Q) = dim H ∗ (MG ; Q). Proof Write A = H ∗ (BG; Q), K its field of fractions and let (A ⊗ H ∗ (M; Q), D) be a semifree model of the Borel fibration (see Proposition 2.109). Then, for degree reasons, D = 0 if and only if the fibration is TNCZ. Therefore, if the fibration is TNCZ, we have by the Borel localization theorem, dim Q H ∗ (M; Q)

= dim K (K ⊗ H ∗ (M; Q)) = dim K H ∗ (K ⊗ H ∗ (M; Q), D) = dim K (K ⊗ H ∗ (MG ; Q)) = dim Q H ∗ (MG ; Q) .

If the fibration is not TNCZ, then we have, in a similar way, dim Q H ∗ (M; Q) > dim K H ∗ (K ⊗ H ∗ (M; Q), D) = dim Q H ∗ (MG ; Q) .

293

294

7 : G-spaces

We now come to an interesting relation between the cohomologies of M and the fixed point set MG when G is a torus. Theorem 7.37 For each n ≥ 0, we have ∞

dim H

n+2i

G

(M ; Q) ≤

i=0

∞

dim H n+2i (M; Q) .

i=0

Proof Once again write A = H ∗ (BG; Q) and let K denote A’s field of fractions. The morphism f : EG ×G MG → EG ×G M is a morphism of fibrations over BG. We construct a morphism of semifree A-modules ϕ : (A ⊗ H ∗ (M; Q), D) → (A ⊗ H ∗ (MG ; Q), 0) inducing H ∗ (f ; Q) in cohomology. As before, let F = MG and also denote BG × F = EG ×G F = FG . Recall that C ∗ (X) is the rational singular cochain algebra on a space X. We denote by

ϕF : C ∗ (BG) ⊗ H ∗ (F; Q) −→ C ∗ (FG ) , and

ϕM : C ∗ (BG) ⊗ H ∗ (M; Q) −→ C ∗ (MG ) semifree models of C ∗ (FG ) and C ∗ (MG ) (see Proposition 2.109). Then, by Proposition 2.107, there exists a morphism of C ∗ (BG)-modules τ such that the following diagram commutes up to homotopy, and therefore in cohomology: C ∗ (M O

ϕM

G)

C ∗ (f )

ϕF

(C ∗ (BG) ⊗ H ∗ (M; Q), d)

/ C ∗ (FG ) O

τ

/ (C ∗ (BG) ⊗ H ∗ (F; Q), d)

Since C ∗ (BG) is related to the minimal model (A, 0) of BG by a sequence of quasi-isomorphisms, there is a morphism of A-semifree modules ϕ : (A ⊗ H ∗ (M; Q), D) → (A ⊗ H ∗ (MG ; Q), 0)

inducing H ∗ (f ; Q) in cohomology. Observe that ϕ A ⊗ H q (M; Q) ⊂ ⊕i≥0 A⊗H q−2i (MG ; Q). We can therefore consider the restricted morphism of complexes ϕ˜ : (A ⊗ H ≤n (M; Q), D) → (A ⊗ H ≤n (MG ; Q), 0) ,

7.4


and the induced morphism of quotient complexes ¯ → (A ⊗ H ≥n (MG ; Q), 0) . ϕ¯ : (A ⊗ H ≥n (M; Q), D) We tensor by K over A and consider the following diagram of complexes (K ⊗ H ≤n (M; Q), D) K⊗A ϕ˜

(K ⊗ H ≤n (MG ; Q), 0)

/ (K ⊗ H ∗ (M; Q), D)

/ (K ⊗ H ≥n (M; Q), D) ¯

K⊗A ϕ

/ (K ⊗ H ∗ (MG ; Q), 0)

K⊗A ϕ¯

q

/ (K ⊗ H ≥n (MG ; Q), 0)

Since q is surjective in cohomology and K ⊗A ϕ is a quasi-isomorphism, K ⊗A ϕ¯ is surjective in cohomology. Note that, for degree reasons, K ⊗A ϕ maps the vector space H n (M; Q) into ⊕i≤0 K ⊗ H n+2i (MG ; Q). This implies the surjectivity in cohomology of the map K ⊗A ϕ¯ : (⊕i≥0 K ⊗ H n+2i (M), D)

/ (⊕ K ⊗ H n+2i (MG ), 0) . i≥0

The result now follows directly.

Theorem 7.7 and Theorem 7.32 are in fact particular cases of a more general result (see [144, Theorem IV.6], [10, definition 4.1.5]): Theorem 7.38 Krull-dim (H ∗ (MG ; Q)) = max {rank Gx | x ∈ X }.

7.4.2

Some examples

Example 7.39 Let M be an even dimensional sphere, M = S2n , with n > 0, and let G = S1 . It follows from Theorems 7.33 and 7.37 that χ (MG ) = 2 and dim H ∗ (MG ; Q) ≤ 2. Therefore, dim H even (MG ) = 2 and dim H odd (MG ) = 0. There are therefore only two possibilities: either MG has the real homotopy type of the union of two points, or else MG is a compact connected manifold that has the rational homology of a sphere and therefore the rational homotopy type of an even dimensional sphere. For instance, fix 0 < p < n and write the sphere S2n as 2 2 p 2(n−p)+1 xj = 1 . (z1 , . . . , zp , x2p+1 , . . . , x2n+1 ) ∈ C × R |zi | + Then the fixed point set of the action of S1 on S2n defined by t(z1 , . . . , zp , x2p+1 , . . . , x2n+1 ) = (tz1 , . . . , tzp , x2p+1 , . . . , x2n+1 ) , t ∈ S1 , is the sphere S2(n−p) .

295

296

7 : G-spaces

Example 7.40 Let M = S2n+1 . If MG = ∅, then we have χ (MG ) = χ (M) = 0 by Theorem 7.33, and dim H ∗ (MG ) ≤ 2 by Theorem 7.37. Thus the fixed point set MG has the rational homotopy type of a sphere S2m+1 with m ≤ n. Example 7.41 The projective space CP(2) is the quotient of C3 \{0} by the equivalence relation identifying the triples (x, y, z) and (λx, λy, λz) for all nonzero λ ∈ C. We consider the action of G = S1 on M = CP(2) given as follows: t[x, y, z] = [tx, y, z] ,

t ∈ S1 .

The fixed point set MG is the union of a point P = [1, 0, 0] and of a submanifold S2 consisting of the elements [0, y, z], y, z ∈ C. The relative minimal model for the Borel fibration has the form (∧x, 0) → (∧x ⊗ ∧(u, v), D) → (∧u, v), d) , with |x| = 2, |u| = 2, |v| = 5, du = 0, dv = u3 . For degree reasons, D(u) = 0 and D(v) = u3 + αxu2 + βx2 u + γ x3 , with α, β, γ ∈ R. The injection of the fixed point P into CP(2) induces a commutative diagram of relative Sullivan models (∧x, 0)

/ (∧x ⊗ ∧(u, v), D)

=

(∧x, 0)

ψ

/ (∧x, 0)

/ (∧u, v), d) / Q

By making a change of generators, we can suppose that ψ(u) = 0. For degree reasons ψ(v) = 0. Since ψ is a morphism of differential graded algebras, we deduce that γ = 0. The minimal model of the injection j of S2 into M is given by ρ : (∧(u, v), d) → (∧(u, v ), d), d(v ) = u2 , ρ(u) = u, ρ(v) = uv . Since the action of S1 on S2 is trivial, j induces a diagram of relative minimal models (∧x, 0)

=

(∧x, 0)

/ (∧x ⊗ ∧(u, v), D)

τ

/ (∧x, 0) ⊗ (∧(u, v ), d)

/ (∧u, v), d)

ρ

/ (∧(u, v ), d)

By making changes of generators, we can suppose that τ (u) = u and τ (v) = α1 xv + uv . The compatibility with the differentials gives β = 0.

7.4


If α = 0, then H ∗ (∧(x, u, v), D) = ∧x ⊗ ∧u/u3 . By the Borel localization theorem, Q(x) ⊗∧x H(∧(x, u, v), D) ∼ = Q(x) ⊗ H ∗ (MG ; Q), so this would ∗ G 3 give H (M ; Q) = ∧u/u , which is not correct. Therefore, α = 0, and by a final change of generators, we can suppose α = 1. Therefore, a minimal model for the Borel fibration associated to the action is given by (∧x, 0) → (∧x ⊗ ∧(u, v), D) → (∧u, v), d) , with |x| = 2, |u| = 2, |v| = 5, D(u) = 0 and D(v) = u3 + xu2 . Note 2 that K ⊗A H ∗ (∧x ⊗ ∧(v, w), D) ∼ = K[t]/t 2 · e1 ⊕ K · e2 with e1 = 1 − ux2 , 2

2

e2 = ux2 , t = u + ux , e1 + e2 = 1. This decomposition corresponds to the decomposition H ∗ (MG ; Q) = H ∗ ({P}; Q) ⊕ H ∗ (S2 ; Q). Example 7.42 Let S1 act on S2 by horizontal rotations. There are two fixed points, the North pole, NP, and the South pole, SP. Let V denote a small neighborhood of NP which is stable under the action. Now denote by S12 , S22 , S32 , S42 four copies of the sphere S2 with the same S1 -action and let V1 , V2 , V3 , V4 denote the corresponding neighborhoods of the North poles. We then have a diagonal S1 -action on S12 × S22 and on S32 × S42 . We remove the interiors of V1 × V2 and of V3 × V4 and paste together the boundaries. We obtain in this way a S1 -action on M = (S12 × S22 )#(S32 × S42 ). The manifold M is rationally hyperbolic, and the fixed point set consists of six isolated points. We verify directly that χ (M) = χ (MG ) = 6 (compare with Theorem 7.33). Example 7.43 Suppose we are given an action of the circle on M = (S13 × S23 )#(S33 × S43 ). Let’s show that the fixed point set is path connected. Since χ (M) = −2, MG is not empty by Theorem 7.33. The vector space H even (MG ; Q) therefore has dimension one or two by Theorem 7.37. In the first case, we would have a manifold MG whose reduced rational cohomology has dimension three and is concentrated in odd degrees. This is not possible by Poincaré duality. We thus have dim H even (MG ) = 2 and dim H odd (MG ) = 4. If there are two components, the sum of the Betti numbers in one component is greater than or equal to 3 and the reduced homology is concentrated in odd degrees. Once again this is not possible by Poincaré duality. Therefore MG is connected and dim H ∗ (MG ) = 6. Example 7.44 We construct an action of S1 on M = (S13 × S23 )#(S33 × S43 ) with fixed point set a torus with two handles. Start with the trivial knot in S3 . The complement is homeomorphic to a product of S1 with an open disk. The rotation along the meridian circles of this solid torus defines an action of S1 on S3 whose fixed point set consists of the original trivial knot. In the next step we choose a closed neighborhood V, stable under the S1 -action,

297

298

7 : G-spaces

of a closed arc contained in the trivial knot. We take four copies of the sphere S3 and of the neighborhood V and we construct the connected sum (S13 × S23 )#(S33 × S43 ) by first removing the interiors of the products V1 × V2 and V3 × V4 and then pasting together their boundaries. The diagonal S1 action on the products then extends to the connected sum, and the fixed point set is a torus with two handles.

7.5

The rational homotopy of a fixed point set component

In this section, we derive information about the rational homotopy Lie algebra LF = π∗ (F) ⊗ Q of a component F of MG from knowledge of the rational homotopy Lie algebra of M, LM = π∗ (M) ⊗ Q. Throughout this section, G denotes a torus which acts smoothly on a compact manifold M. Note that the application of rational homotopy is not easy here because MG is not necessarily path connected and therefore does not admit a minimal model. The essential part of this section consists of a presentation of theorems due to Allday and Puppe. We will use Z/2Z-graded Sullivan models in the proof of the main theorem. For the sake of simplicity, we have concentrated all the algebraic ingredients necessary for the proof in a subsection. The reader who is only interested in topological applications can avoid reading that subsection.

7.5.1

The rational homotopy groups of a component

G G Thefixedpoint set M decomposes as a finite union of submanifolds M = F1 F2 . . . Fn . The component F1 is denoted F. The injection of F into M induces a commutative diagram of Borel fibrations and a commutative diagram of corresponding relative minimal models:

F

/ FG

/ BG

M

/ (A ⊗ ∧XM , D)

A

A

/ MG

=

f

/ (A, 0) ⊗ (∧XF , d)

=

/ BG / (∧XM , d) / (∧XF , d)

Here as usual A = H ∗ (BG; Q) and K is its field of fractions.

7.5


Theorem 7.45 ([3]) With the notation above, there exists a differential D on the Z/2Z-vector space K⊗XM such that the cohomology H ∗ (K⊗XM , D) is isomorphic to K ⊗ XF as a Z/2Z-vector space. Proof The canonical augmentation ε : (∧XF , d) → (Q, 0) extends naturally to an A-linear map ε : (A, 0) ⊗ (∧XF , d) → (A, 0). We tensor ε by K over A and denote by I the kernel of the augmentation map ε

f

: (K ⊗ ∧XM , D) → (K, 0) ⊗ (∧XF , d) → (K, 0) . Now let xi be a basis of XM indexed by a well ordered set such that d(xi ) ∈ A ⊗ ∧(xj , j < i). For each xi , the element x˜ i = xi − (xi ) belongs ˜ M the vector space to I, with (xi ) = 0 when |xi | is odd. We denote by X generated by the x˜ i . The isomorphism ˜M θ : K ⊗ ∧XM → K ⊗ ∧X ˜ M. defined by θ (xi ) = x˜ i +(xi ) induces a differential D = θDθ −1 on K⊗∧X ˜ Note that the differential algebra (K ⊗ ∧XM , D) is a Z/2Z-graded Sullivan algebra, but not necessarily a minimal one. Write Q(D) for the differential induced on the indecomposables. ˜ M ) ⊂ K ⊗ ∧+ XF , and f induces a map on Since f (I) ⊂ K ⊗ ∧+ XF , f (X the homology of indecomposables ˜ M , Q(D)) −→ K ⊗ XF . Q(f ) : H(K ⊗ X We will now prove that this map is an isomorphism and this will imply the result. By the localization theorem we have an isomorphism ∗ ∗ (M; Q) → K ⊗A HG (MG ; Q) θ : K ⊗A HG

∼ = ⊕ni=1 (K ⊗ H ∗ (Fi ; Q)) . = K ⊗ H ∗ (MG ; Q) ∼ If we write H 0 (Fi ; Q) = Qei with ei2 = ei , then H 0 (MG ; Q) is the vector space generated by the orthogonal idempotents ei : ei2 = ei ,

ei · ej = 0 if i = j .

∗ (M; Q) such that θ (x) = e . By inverting We choose an element x in K⊗A HG 1 the element θ (x), we kill the components H ∗ (Fi ; Q) for i ≥ 2. We get an isomorphism of algebras

∗ θ¯ : K ⊗A HG (M; Q) [x−1 ] −→ K ⊗ H ∗ (MG ; Q) [θ (x)−1 ]

∼ = K ⊗ H ∗ (F; Q) .

299

300

7 : G-spaces

∼ K ⊗A H ∗ (M; Q) shows that ˜ M , D) = Now, the isomorphism H ∗ (K ⊗ ∧X G ˜ M such that [u] = x. We then consider there is a cocycle u in K ⊗ ∧X ˜ M ⊗ ∧(v, w), D), where the Z/2Z-graded relative minimal model (K ⊗ ∧X |v| = 0 , |w| = 1 , D(v) = 0 , and D(w) = uv − 1. ˜ M , D)[x−1 ] → H ∗ (K ⊗ ∧X ˜M ⊗ Define a morphism ψ : H ∗ (K ⊗ ∧X −1 ∧(v, w), D) by putting ψ(x ) = [v]. Recall that a localized module B[x−1 ] is by definition the quotient B[x−1 ] = B ⊗ ∧y /(xy − 1). In our case, since ˜ M ⊗ ∧(v, w), D), the morphism ψ is well defined uv − 1 = 0 in H ∗ (K ⊗ ∧X and is an isomorphism by Lemma 7.46. We extend the map f : (K ⊗ ∧XM , D) → (K, 0) ⊗ (∧XF , d) to a map ˜ M ⊗ ∧(v, w), D) → (K, 0) ⊗ (∧XF , d) , f¯ : (K ⊗ ∧X by defining f¯ (v) = 1 and f¯ (w) = 0.

∼ =

The Borel localization isomorphism H ∗ (K ⊗ ∧XM )[x−1 ] −→ K ⊗ H ∗ (∧XF , d) then factors as the composition ψ ˜ M )[x−1 ] −−− ˜ M ⊗ ∧(v, w), D) −→ H ∗ (K ⊗ ∧X H ∗ (K ⊗ ∧X ∼ =

−H ∗ (f¯ )

−−−−→ K ⊗ H ∗ (∧XF , d). Since ψ is an isomorphism, H ∗ (f¯ ) is also an isomorphism. We introduce the elements u˜ = u − 1 and v˜ = v − 1 that belong to the kernel of ε ◦ f¯ , and we have D(w) = u˜ + v˜ + u˜ ˜ v. In the same way as in the classical theory of N-graded Sullivan models, an isomorphism between Z/2Z-graded Sullivan models induces an isomorphism on the homology of indecomposable elements (see Lemma 7.56). ˜M ⊕ ˜ M , Q(D)) ∼ In our case this implies that H(K ⊗ X = H(K ⊗ (X ∼ ˜ (˜v, w)), Q(D)) = K ⊗ XF . Since K ⊗ XM is isomorphic to K ⊗ XM as a graded vector space, we obtain a differential D on K ⊗ XM such that ˜ M , Q(D)). H(K ⊗ XM , D) ∼ = H(K ⊗ X Lemma 7.46 Let (B ⊗ ∧(v, w), D) be a differential Z/2Z-graded commutative algebra with |v| = 0, |w| = 1, D(v) = 0 and D(w) = uv − 1, where u is a cocycle of degree 0 in B. Then the morphism ψ : H ∗ (B)[u−1 ] → H ∗ (B ⊗ ∧(v, w), D) defined by ψ(u−1 ) = [v] is an isomorphism. Proof We filter the cochain algebra (B⊗∧(v, w), D) by the degree in B⊗∧v. We get a spectral sequence whose E2 -term satisfies

0,∗ 1,∗ 0,∗ E2 = E2 ⊕ E2 , E2 = H ∗ (B, d) ⊗ ∧v /(uv − 1) = H ∗ (B, d)[u−1 ] . n 1,∗ i A cocycle α in E2 decomposes into a sum α = i=r bi v w, with bi ∈ ∗ H (B, d). The condition d1 (α) = 0 implies that br = 0. Therefore, by

7.5


1,∗

induction, we have E2 = 0. The spectral sequence therefore collapses at the E2 -level and this shows that ψ is an isomorphism. Here is a main consequence of Theorem 7.45. It gives a powerful connection between a G-manifold and its fixed point set when the extra condition of rational ellipticity is added. Corollary 7.47 ([3]) Let G be a torus, M a compact nilpotent rationally elliptic G-manifold and F a nilpotent component of MG . Then 1. F is a rationally elliptic manifold; 2. χπ (F) = χπ (M); 3. dim π∗ (F) ⊗ Q ≤ dim π∗ (M) ⊗ Q. Proof With the above notation, by Theorem 7.45, dim XF = dim K H ∗ (K ⊗ ˜ M ), XM , D), so clearly we have dim XF ≤ dim XM . If we let χ (XM ), χ (X χ (XF ) denote the Euler–Poincaré characteristics of the respective complexes ˜ M , K ⊗ XF , the standard result that the Euler–Poincaré K ⊗ XM , K ⊗ X characteristic is invariant under taking homology gives ˜ M ) = χ (XF ) = χπ (F) . χπ (M) = χ(XM ) = χ(X

As a corollary we recover a special case of the well-known theorem that the inclusion of an orbit for an effective torus action on an aspherical manifold induces an injection at the fundamental group level (see [63]). Theorem 7.48 Let G be a torus acting on a nilmanifold M = K(π , 1). 1. If the action is nontrivial, then MG = ∅. 2. If the action is effective, then the injection of an orbit G·x → M induces an injective map π1 (G) → π. Proof The minimal model of a nilmanifold M has the form (∧XM , d), where XM is a finite dimensional vector space concentrated in degree one. Moreover, the dimension of XM is equal to the dimension of the manifold M (see Remark 3.21). Suppose MG = ∅ and let F be a path component of MG , and (∧XF , d) its minimal model. By Theorem 7.45, H ∗ (K ⊗ XM , D) ∼ = K ⊗ XF . But XM is concentrated in degree 1, so K ⊗ XM is in odd degrees only. Thus, we have D = 0 and, therefore, K ⊗ XM ∼ = K ⊗ XF as a Z/2Z-graded vector space. Thus, the dimension of XF is equal to the dimension of XM . Now, XF is concentrated in odd degrees, but if there is an element in an odd degree larger than 1, then the dimension of F would exceed the dimension of M by Theorem 2.75 (6). Thus, XF is concentrated in degree one. This implies that

301

302

7 : G-spaces

F is a closed submanifold of the same dimension as M, so MG = F = M, which is not possible. Therefore MG = ∅. Suppose now that the action is effective. We shall show that the action is, in fact, almost free. If not, then there exists some S1 in some isotropy group at a point x ∈ M, say. Since we assume the action is effective, we know that this S1 does not act trivially on the whole manifold M. By (1), we also know 1 that MS = ∅, but this contradicts the fact that S1 is in the isotropy group of x. Hence, no isotropy group contains an S1 and the action is almost ∗ (M; Q) is finite free. By Hsiang’s theorem, Theorem 7.7, we have that HG dimensional. Now let ¯ (∧(x1 , . . . , xr ), 0) → (∧(x1 , . . . , xr , y1 , . . . , yn ), D) → (∧(y1 , . . . , yn ), D) be a relative minimal model for the Borel fibration M → MG → BG, with G = T r . We have, as usual, |xi | = 2 and, because M is a nilmanifold, |yi | = 1. Because (∧(x1 , . . . , xr ), d = 0) is a polynomial algebra and all the other generators of (∧(x1 , . . . , xr , y1 , . . . , yn ), D) are in degree 1, the only way to achieve finite dimensionality for cohomology is to kill the entire vector space in degree 2 having basis {x1 , . . . , xr }. Therefore, there must exist r elements yi1 , . . . , yir such that the elements D(yij ) form a basis for the vector space generated by the xk . In particular, this says that the D(yij ) all have nonzero linear parts. Recall from Proposition 2.65 that the linear part of a differential in a relative model represents the connecting map in the long exact homotopy sequence of the associated fibration. Here this means that the connecting map π2 (BG) ⊗ Q → π1 (M) ⊗ Q is injective (since the entire vector space generated by the xk ’s is hit by the differential D). Since π2 (BG) is torsion-free, the usual connecting map π2 (BG) → π1 (M) is also injective. Recalling that the composition π1 (G) = π1 (BG) ∼ = π2 (BG) → π1 (M) may be identified with the map induced by the injection of an orbit, we obtain the result. Corollary 7.49 Suppose that G is an r-torus acting on a compact nilpotent rationally elliptic manifold M with χπ (M) = 0. Then MG has no contractible component. Proof For each component F of MG , XF = 0 and so, by Corollary 7.47, the rational cohomology of F is not trivial. Example 7.50 Let G be a torus acting on the projective n-space M = CP(n). Since the rational cohomology of M is zero in odd degrees, by Theorem 7.37, the cohomology of MG is concentrated in even degrees and dim H ∗ (MG ; Q) = n + 1. Denote by F a simply connected component of MG . Since χπ (F) = χπ (M) = 0, and dim π∗ (F)⊗ Q ≤ dim π∗ (M)⊗ Q = 2, F

7.5


has the rational homotopy type of a point or of a complex projective space CP(m) with m ≤ n. Example 7.51 Let G be a torus acting on a nilpotent rationally hyperbolic manifold M, and suppose that πeven (M) ⊗ Q is finite dimensional. Let F be a component of MG . Then with the notation of Theorem 7.45, H ∗ (K⊗XM , D) is infinite dimensional, and so, by Theorem 7.45, K ⊗ XF ∼ = H ∗ (K ⊗ XM , D) is also infinite dimensional. In particular, each nilpotent component of the fixed point set is a rationally hyperbolic manifold. Example 7.52 Let M be the connected sum of two copies of the product S3 × S3 × S3 × S3 and let G be a torus acting on M. Denote by z ∈ π∗ (M) ⊗ Q the class of the image of the boundary sphere in (S3 )4 \{∗}, and by y ∈ π∗ (M)⊗ Q the image of z by adjunction. The construction of the differential in the minimal model of the connected sum (see Example 3.6) shows that the element y is a generator in the Lie algebra π∗ (M) ⊗ Q. Therefore, by Corollary 3.4, the rational homotopy Lie algebra π∗ (M) ⊗ Q is a graded Lie algebra generated in degrees 2 and 10. It follows from Example 7.51 that π∗ (F) ⊗ Q is also infinite dimensional.

7.5.2

Presentation of the Lie algebra LF = π∗ (F) ⊗ Q

When πeven (M) ⊗ Q = 0, or more generally, when the differential D defined in Theorem 7.45 on the Z/2Z-graded vector space K ⊗ XM vanishes, the Borel localization theorem gives information about the Lie algebra π∗ (F) ⊗ Q. Theorem 7.53 ([9]) Suppose M is a compact smooth T r -manifold satisfying πeven (M) ⊗ Q = 0, and let F be a nilpotent component of MG . Denote by

L(WM ) → L(GM ) → LM and

L(WF ) → L(GF ) → LF minimal presentations, where L(−) denotes the graded free Lie algebra functor. Then, 1. dim GF ≤ dim GM ; 2. dim WF ≤ dim WM . Proof Since the action is smooth, for every generic circle S1 ⊂ T r , we have 1 r X S = X T (see Lemma 7.3). Thus we may assume without loss of generality that r = 1. In this case, A = Q[t] and K = Q(t). For α ∈ Q, let Qα indicate that Q has been made into an A module via the map A → Q given by t → α.

303

304

7 : G-spaces

We use the notation contained in the proof of Theorem 7.45. Since XM ˜ M = XM and the differential Q(D) on the is concentrated in odd degrees, X indecomposable elements is zero. Therefore f induces an isomorphism of Z/2Z-graded differential algebras K ⊗A (A ⊗ ∧XM , D1 ) → (K, 0) ⊗ (∧XF , d1 ) , where D1 (XM ) ⊂ A⊗∧2 XM and d1 is the quadratic part of the differential d. On the other hand, by the definition of the relative minimal model of MG , we have Q0 ⊗A (A⊗∧XM , D1 ) = (∧XM , d1 ). Denote by C ∗ the complex (A⊗ ∧∗ XM , D1 ). The universal coefficient theorem gives a short exact sequence i+1 0 → Q0 ⊗A H i (C ∗ ) → H i (Q0 ⊗A C ∗ ) → TorA (C ∗ ), Q0 ) → 0 . 1 (H

Note, in particular, that the indices refer to the gradation A ⊗ ∧∗ XM . Now observe that, for a morphism ϕ : P1 → P2 of A-modules, K ⊗A ϕ is an isomorphism if and only if Q1 ⊗A ϕ is an isomorphism. Then, using Lemma 7.54 below, we have the following sequence of equalities. dim Q H q (∧XF , d1 ) = dim Q Q1 ⊗A H q (A ⊗ ∧XF , d1 ) = dim Q Q1 ⊗A H q (A ⊗ ∧XM , D1 ) q ∗ = dim Q Q0 ⊗A H q (C ∗ ) − dim Q TorA 1 (H (C ), Q0 ) q+1 (C ∗ ), Q0 ) = dim Q H q (Q0 ⊗A C ∗ ) − dim Q TorA 1 (H q ∗ − dim Q TorA 1 (H (C ), Q0 ) q+1 (C ∗ ), Q0 ) = dim Q H q (∧XM , d1 ) − dim Q TorA 1 (H q ∗ − dim Q TorA 1 (H (C ), Q0 ).

By Subsection 2.6.2, we have WM ∼ = H 2 (∧XM , d1 ), WF ∼ = H 2 (∧XF , d1 ), 1 1 ∼ GM = H (∧XM , d1 ) and GF = H (∧XF , d1 ). The result then follows directly from the equalities above. Lemma 7.54 Let P be a graded A-module and suppose that dim (Q0 ⊗A P) < ∞. Then dim Q (Q0 ⊗A P) = dim Q (Q1 ⊗A P) + dim Q TorA 1 (P, Q0 ) . Proof Since dim (Q0 ⊗A P) < ∞, P is finitely generated, and we can write P as the direct sum of a free part and submodules of the form Q[t]/(t n ). Let 0 → N → L → P → 0 be a minimal free presentation of P. We directly deduce the following exact sequences: 0 → TorA 1 (P, Q0 ) → Q0 ⊗A N → Q0 ⊗A L → Q0 ⊗A P → 0

7.5


and 0 → Q1 ⊗A N → Q1 ⊗A L → Q1 ⊗A P → 0 .

7.5.3

Z/2Z-Sullivan models

The purpose of this section is to extend to Z/2Z-graded Sullivan algebras the basic properties of usual Sullivan algebras. All the material of this section is due to C. Allday [3]. In what follows all vector spaces will be defined over a field lk of characteristic zero. A Z/2Z-graded vector space A is a vector space with a decomposition A = A0 ⊕ A1 . It is a Z/2Z-graded algebra if there is a multiplication satisfying Ap · Aq ⊂ Ap+q , where the addition p + q is taken modulo 2. The algebra A is called commutative if for homogeneous elements a and b, we have a·b = (−1)|a|·|b| b·a. It is called free if it is the tensor product of a polynomial algebra on elements of degree 0 with an exterior algebra on elements of degree 1; in this case we write A = ∧V where V is the linear span of the generators of the algebra. A Z/2Z-graded differential algebra (A, d) is a Z/2Z-graded algebra A equipped with a differential d : Ap → Ap+1 . For instance, each cochain algebra (A, d) defines a Z/2Z-graded differential algebra with A0 = Aeven and A1 = Aodd . A Z/2Z-graded Sullivan algebra is a Z/2Z-graded differential algebra (A, d) such that A is the free commutative graded algebra on a vector space V that admits a basis indexed by a well-ordered set (xi ) with d(xi ) ∈ ∧(xj , j < i). We write (A, d) = (∧V, d). It is called a Z/2Z-graded minimal Sullivan algebra if d(V) ⊂ ∧≥2 V. For instance every Sullivan algebra induces a Z/2Z-graded Sullivan algebra and a minimal Sullivan algebra (∧V, d) induces a Z/2Z-graded minimal Sullivan algebra. A morphism of Z/2Z-graded differential algebras f : (A, d) → (B, d) is a degree zero linear map that is compatible with the multiplications (f (ab) = f (a)f (b)) and with the differentials (df = fd). As usual, a quasi-isomorphism is a morphism that induces an isomorphism in cohomology. Let (A, d) be a Z/2Z-graded differential algebra. An augmentation is a surjective map of differential graded algebras ε : (A, d) → (lk, 0), where lk is in degree 0. A Z/2Z-graded Sullivan algebra (∧V, d) is equipped with a natural augmentation sending V to 0. A morphism between Z/2Z-graded Sullivan algebras f : (∧V, d) → (∧W, d) preserves the augmentations if f (V) ⊂ ∧+ W. When (∧V, d) is a Z/2Z-graded Sullivan algebra, the vector space V is called the vector space of indecomposable elements. The differential d

305

306

7 : G-spaces

induces a differential Q(d) on V defined as the composition p d Q(d) : V −→ ∧+ V −→ ∧+ V/ ∧≥2 V ∼ =V.

Here, p denotes the canonical projection. The homology H ∗ (V, Q(d)) is called the homology of indecomposable elements. An augmentation preserving map f : (∧V, d) → (∧W, d) induces a map of complexes Q(f ) : (V, Q(d)) → (W, Q(d)). To define homotopies, we consider the contractible Sullivan algebra (∧(t, dt), d) with |t| = 0, |dt| = 1 and d(t) = dt. We denote by pi : (∧(t, dt), d) → (lk, 0), i = 0, 1, the augmentations defined by pi (t) = i and pi (dt) = 0. Definition 7.55 The morphisms f , g : (A, d) → (B, d) are homotopic if there exists a morphism H : (A, d) → (B, d) ⊗ ∧(t, dt) such that f = (1 ⊗ p0 ) ◦ H and g = (1 ⊗ p1 ) ◦ H. Lemma 7.56 ([3]) An augmentation preserving quasi-isomorphism f : (A, d) → (B, d) between Z/2Z-graded Sullivan algebras induces an isomorphism on the homology of indecomposable elements.

7.6

Hamiltonian actions and bundles

The symmetries which are important in symplectic geometry all involve the symplectic form. Although actions of more general compact Lie groups on symplectic manifolds may be considered, here we shall concentrate on circle and torus actions to avoid the machinery of moment maps.

7.6.1

Basic definitions and properties

From now on, (M2n , ω) denotes a closed symplectic manifold. Definition 7.57 • A smooth action of the circle, A : S1 × M → M is symplectic if g ∗ ω = ω

for each g ∈ S1 . • An orbit map α : S1 → M is defined by choosing x ∈ M and taking the orbit through x; g → gx. We may consider α ∈ π1 (M) and we denote its Hurewicz image by h(α) ∈ H1 (M; R). Note that all orbit maps induce the same map on homology (since M is always connected). • The fundamental vector field X associated to the action is defined at each x ∈ M as Dx α(∂/∂θ), where ∂/∂θ is the unit vector field of S1 and Dx α is the induced map on tangent spaces.

7.6


The main point that is elicited by the definition of symplectic action is that the Lie derivative L(X) (see Section A.2) vanishes on forms that are invariant under the action (compare Section 1.7, especially Proposition 1.33). This property applied to the symplectic form ω gives 0 = L(X)ω = (di(X) + i(X)d)ω = di(X)ω, where i(X) denotes interior multiplication by X (see Section A.2). Here we use the formula in Proposition A.4 and the fact that dω = 0. Therefore, the form i(X)ω is a closed form. Of course, the same holds for every invariant form on M. Moreover, by Theorem 1.28, we know that the cohomology of the invariant forms is the same as the ordinary de Rham cohomology, so we can consider i(X) as a derivation on H ∗ (M; R). To see this, let β and β be invariant forms that represent the same cohomology class. Hence, there is an invariant form γ with β − β = dγ . But then, i(X)β − i(X)β = i(X)dγ = d(−i(X)γ ) since L(X)γ = 0. Therefore, in H ∗ (M; R), [i(X)β] = [i(X)β ]. Definition 7.58 The λ-derivation on H ∗ (M; R) is defined by λ(b) = [i(X)β], where b ∈ H ∗ (M; R) is represented by the invariant form β. Proposition 7.59 ([179]) The λ-derivation may be identified directly from the homomorphism induced on cohomology by the action A; A∗ b = 1 × b + uS1 × λ(b), where uS1 is the generator of H 1 (S1 ; R) corresponding to the unit 1-form dθ on S1 . In particular, when b ∈ H 1 (M; R), λ(b) = α ∗ (b), [S1 ]. So now we have two ways to view the λ-derivation. In fact, the λderivation has already appeared in the proof of Theorem 4.98. Recall from Definition 4.96 that (M2n , ω) has Lefschetz type if multiplication by [ω]n−1 ∼ =

is an isomorphism: Lef = ωn−1 ∪ : H 1 (M) → H 2n−1 (M). For the sake of simplicity, we denote the symplectic form and its cohomology class by ω. Now here are a few properties of the λ-invariant, see [180]. Proposition 7.60 With the notation of Definition 7.57 and the preceding discussion, we have: 1. h(α) is Poincaré dual to nλ(ω) ωn−1 = n Lef(λ(ω)); 2. λ(ωn ) = 0 if and only if λ(H 1 (M)) = 0 if and only if h(α) = 0;

307

308

7 : G-spaces

3. If (M2n , ω) has Lefschetz type, then the three properties of part 2 are equivalent to λ(ω) = 0. Proof (1) Let b ∈ H 1 (M) and note that b ωn = 0 for degree reasons. Applying λ, we have 0 = λ(b)ωn − bλ(ωn ) = λ(b)ωn − b nλ(ω)ωn−1 using the fact that λ is a graded derivation. Hence, λ(b)ωn = b nλ(ω)ωn−1 . Taking the Kronecker product with the fundamental class of M gives bλ(ωn ), [M] = λ(b)ωn , [M] = λ(b). But λ(b) = α ∗ (b), [S1 ] = b, α∗ [S1 ] = b, h(α), so b, h(α) = λ(b) = bλ(ωn ), [M] = b nλ(ω)ωn−1 , [M] = b, nλ(ω)ωn−1 ∩ [M] and we see that h(α) is Poincaré dual to nλ(ω)ωn−1 . (2) The first equivalence follows exactly as in the proof of Theorem 4.36. The second equivalence comes from λ(b) = b, h(α) for all b ∈ H 1 (M). (3) Suppose M has Lefschetz type. If λ(ω) = 0, then it is clear that λ(ωn ) = 0. If λ(ωn ) = 0, then we have 0 = λ(ωn ) = nλ(ω)ωn−1 = nLef(λ(ω)). But the Lefschetz type assumption then implies that λ(ω) = 0.

7.6.2

Hamiltonian and cohomologically free actions

The importance of Proposition 7.60 comes from the following fundamental definitions in symplectic geometry. If A : T k × M → M is a symplectic toral action, then we may define k λ-derivations by λj (ω) = [i(Xj )ω], where Xj is the fundamental vector field of the jth circle factor in T k . Definition 7.61 For a symplectic manifold (M2n , ω): 1. a symplectic action A : T k × M → M is Hamiltonian if λj (ω) = 0 for each j = 1, . . . , k. In particular, if H 1 (M; R) = 0, then all symplectic actions are Hamiltonian; 2. a symplectic action A : T k × M → M is cohomologically free if the homomorphism on cohomology, ψ : H 1 (T k ; R) → H 1 (M; R), defined by ψ(uS1 ) = λj (ω) j

is injective. Here uS1 denotes the generator of H 1 (T k ; R) corresponding j

to the j-th circle factor.

7.6


Remark 7.62 Even if (M2n , ω) is only c-symplectic with c-symplectic cohomology class ω, we can, by Proposition 7.59, still define λ(ω) for any S1 -action and say that the action is c-Hamiltonian if λ(ω) = 0. Similarly, we can define a notion of cohomologically free for c-symplectic manifolds. See [179] for details. Recall from Theorem 1.102 that, for an S1 -action on a closed manifold M, the Cartan–Weil model is the complex (X [u], dX ), where X is the sub-complex of ADR (M) consisting of S1 -invariant forms, u is a degree 2 generator and the differential is defined by dX u = 0, and dX α = dα − ui(X)α, for α ∈ X . The Cartan–Weil model induces an isomorphism H ∗ (X [u], dX ) ∼ = H ∗ (MS1 ; R). Thus, the Cartan–Weil model calculates the equivariant cohomology associated to the action of S1 on M. The equation dX α = dα − ui(X)α says something interesting. Although α is an invariant form, even if it is closed, this does not mean that α is equivariantly closed (i.e. dX -closed). For that, we also require i(X)α = 0. Now suppose the S1 -action is a symplectic action so that the symplectic form ω is an invariant form. Then we have dX ω = dω − ui(X)ω = −ui(X)ω since dω = 0. Therefore, the interior multiplication i(X)ω is the obstruction to ω being equivariantly closed: that is, representing a cohomology class in HS∗1 (M; R). Of course, we might be able to change ω within its cohomology class in (X [u], dX ) and achieve equivariant closure. So let’s ask; how can we augment ω in (X [u], dX ) to make it equivariantly closed? The only possibility, for degree reasons, is to take ω = ω − fu for some smooth function f : M → R. The definition of dX then gives dX ω = dX ω − dX (fu) = dω − ui(X)ω − (dX f )u − fdX u = −ui(X)ω − (dX f )u = −ui(X)ω − (df − ui(X)f )u = −ui(X)ω − (df )u = −u(df + i(X)ω). Note that ω is closed exactly when df + i(X)ω = 0. Since λ(ω) = [i(X)ω], this is exactly when the action is Hamiltonian. Thus, denoting the Borel i

fibration of the S1 -action by M → MS1 → BS1 , we have

309

310

7 : G-spaces

Theorem 7.63 ([15]) A symplectic S1 -action on (M2n , ω) is Hamiltonian if and only if ω ∈ H 2 (M; R) is in the image of i∗ : H 2 (MS1 ; R) → H 2 (M; R). Now, the pullback of the Borel fibration M → MS1 → BS1 via the inclusion S2 → BS1 gives a fibration M → E → S2 and this fibration has an associated Wang sequence: i∗

λ

· · · → H 2 (E) → H 2 (M) → H 1 (M) → · · · ∼ H 2 (M 1 ). It may be shown where λ is the Wang derivation and H 2 (E) = S (see [180] and Proposition 7.59) that this Wang derivation is precisely the derivation defined in Definition 7.58. This then gives another “formless” proof of Theorem 7.63 since exactness of the Wang sequence implies that λ(ω) = 0 holds precisely when ω ∈ Im(i∗ ). This Wang approach indicates how algebraic topological methods can often take the place of geometric ones and, in the process extend the relevant notions. Now let’s return to consideration of the properties of Hamiltonian actions. In particular, we are interested in the fixed point set. Proposition 7.64 A Hamiltonian action on a compact symplectic manifold has at least two components in its fixed point set. If the manifold has Lefschetz type, then the existence of a fixed point implies that the action is Hamiltonian. Proof We give the proof for S1 because fixed point questions always reduce to this case (see Lemma 7.3). If λ(ω) = [i(X)ω] = 0, then i(X)ω = dH for some smooth function H : M → R. Since M is compact, H has a maximum y and minimum z on M, so 0 = dHy = i(Xy )ω and 0 = dHz = i(Xz )ω. But ω is nondegenerate, so we can only have these equalities if Xy = 0 = Xz . Since a zero for the fundamental vector field corresponds to a fixed point for the action, we are done. If M has Lefschetz type (see Definition 4.96), then the orbit map at a fixed point gives h(α) = 0 and (3) of Proposition 7.60 then says that the action is Hamiltonian. Remark 7.65 Note by [179] that this theorem also holds in the case of c-symplectic manifolds with c-Hamiltonian action. Proposition 7.66 Any symplectic torus action A : T k × M → M splits as T k = T r × T k−r , where the action restricted to T r is Hamiltonian and the action restricted to T k−r is cohomologically free. For a proof of this result, see [162] and [179] for a cohomological version. These types of splittings have long been used in transformation groups. For instance, the action of a torus can be split into one with finite isotropy

7.6


groups and one with fixed points. Now let’s characterize cohomologically free actions in the case of Lefschetz type. Theorem 7.67 ([179]) Suppose that (M2n , ω) has Lefschetz type and that A : T k × M → M is a symplectic torus action with orbit map α : T k → M. Then 1. The action is cohomologically free if and only if the induced homomorphism α∗ : H1 (T k ; R) → H1 (M; R) is injective. 2. The action is cohomologically free if and only if it is almost free. Proof In the following, we take R-coefficients in cohomology. (1) Suppose the action is cohomologically free and choose a vector space splitting s : H 1 (M) → H 1 (T k ) for ψ : H 1 (T k ) → H 1 (M) (since ψ is injective). Let ∼ =

PD : H1 (M) → H 2n−1 (M) denote the Poincaré duality isomorphism. Then for each S1 factor in T k , we have the following (using Proposition 7.60 (1)). (Note that multiplication by ωn−1 , denoted Lef, is an isomorphism by the Lefschetz type assumption; see Definition 4.96.) s(1/n Lef −1 PDα∗ ([S1 ])) = s(1/n Lef −1 PD(h(α))) = s(1/n Lef −1 (nλ(ω)ωn−1 )) = sλ(ω) = sψ(uS1 ) = uS1 . The composition H1 (T k ) → H 1 (T k ) is then the Poincaré duality isomorphism, so α∗ must be injective. On the other hand, suppose α∗ is injective and choose a vector space splitting t : H1 (M) → H1 (T k ). Then we can compute. tPD−1 (n Lef(ψ(uS1 ))) = tPD−1 (n Lef(λ(ω))) = tPD−1 (n λ(ω)ωn−1 ) = t(h(α)) = t(α∗ ([S1 ])) = [S1 ]. Again, since this expresses the Poincaré duality isomorphism, we see that ψ is injective and the action is cohomologically free. (2) If the action is cohomologically free, then α∗ is injective by (1). Suppose some isotropy group is not finite; thus, there is a subtorus T s which fixes a point x. The orbit map at x then induces a homology factorization α∗ : H1 (T k ) → H1 (T k /T s ) → H1 (M). But the first homomorphism is

311

312

7 : G-spaces

clearly not injective, so α∗ is not injective. Hence, all isotropy groups are finite and the action is almost free. Suppose the action is almost free and write T k = T r × T k−r as in Proposition 7.66. By Proposition 7.64, since the T r action is Hamiltonian, it has fixed points. Because the action is almost free, we must have T r = ∗ and T k acts cohomologically freely.

7.6.3

The symplectic toral rank theorem

In Subsection 7.3.3 we discussed the far-reaching toral rank conjecture, which says that the size of a space’s cohomology limits its ability to admit large almost free toral actions. In our context, we ask how large the cohomology of a symplectic manifold must be in order to admit a symplectic T k action. We can answer this question in the case of Lefschetz type. Theorem 7.68 If (M2n , ω) is a closed symplectic manifold of Lefschetz type and T k acts symplectically and almost freely on M, then dim H ∗ (M; R) ≥ 2k . Proof By (2) and then (1) of Theorem 7.67, we have that α∗ : H1 (T k ) → H1 (M) is injective. Take the Borel fibration (see Section 7.2) associated to the action and go one step back in the associated Puppe sequence (using BT k T k ) to get α

T k → M → MT k . Because α∗ is injective, duality says that α ∗ : H 1 (M) → H 1 (T k ) is surjective. Now, since H ∗ (T k ) is generated by H 1 (T k ), we see that α ∗ : H ∗ (M) → H ∗ (T k ) is surjective as well. Therefore, by Definition 4.39, the fibration is totally noncohomologous to zero (i.e. TNCZ) and we have an isomorphism H ∗ (M) ∼ = H ∗ (T k ) ⊗ H ∗ (MT k ). Thus, dim H ∗ (M) ≥ dim H ∗ (T k ) = 2k .

Remark 7.69 In Theorem 7.24, we used models to show that the inequality holds for any c-symplectic manifold whose cohomology satisfies the hard Lefschetz property (see Theorem 4.35). Here we have given a more traditional approach using only the weaker Lefschetz type property (Definition 4.96).

7.6.4

Some properties of Hamiltonian actions

When a torus action is Hamiltonian, there are interesting (and restrictive) consequences beyond those mentioned above (i.e. existence of fixed points).

7.6


One of the most famous is the following (see, for instance, [161, Corollary 4.8]). Theorem 7.70 Let T k act on M and let M → MT k → BT k be the associated Borel fibration. If the T k action on (M2n , ω) is Hamiltonian, then, as vector spaces, H ∗ (MT k ; R) ∼ = H ∗ (M; R) ⊗ H ∗ (BT k ; R). Of course, we see that this is equivalent to the fibration being TNCZ. Here is a situation where the symplectic and c-symplectic worlds diverge. Example 7.71 ([5]) By Theorem 7.36, we know that the Borel fibration X → XG → BG is TNCZ if and only if dim H ∗ (X G ; Q) ∼ = dim H ∗ (X; Q) where X G is the fixed point set. Now let S1 act freely on S3 × S3 by translation on the first factor. Take an equivariant tube C = S1 × D5 about an orbit (via the slice theorem) and note that ∂C = S1 × S4 = ∂(D2 × S4 ). Let X = S3 × S3 − C and sew in D2 × S4 to obtain N = X ∪∂C (D2 × S4 ). We let S1 act by rotations on D2 and we observe that this is compatible with the action by translations of S1 on itself (i.e. the action in S3 ). Of course, the center of D2 is fixed by this action, so we see that S1 acts on N semifreely (i.e. isotropy is either zero or 1 all of S1 ) with N S = S4 . Note that N has the following cohomology. ⎧ ⎪ j = 1, 5 ⎨0 j H (N; Z) = Z j = 2, 4, 6 ⎪ ⎩ Z ⊕ Z j = 3. Now let θ [z0 , z1 , z2 ] = [eiθ z0 , z1 , z2 ] define a semifree action of S1 on CP(3) with fixed set p ∪ CP(2). Note that p = [1, 0, 0] is fixed by the homogeneity of the coordinates. The fixed CP(2) consists of all points [0, z1 , z2 ]. Since CP(3) and N have fixed points, we can take small equivariant disks about chosen fixed points and form the equivariant connected sum M = CP(3)#N. The fixed set of the S1 action on M is given by 1 MS = p ∪ (CP(2)#S4 ) = p ∪ CP(2). Now, M is clearly c-symplectic since the generator of H 2 (CP(3)) multiplies to a top class. Furthermore, the action is c-Hamiltonian since H 1 (CP(3); Z) = 0 and H 1 (N; Z) = 0 imply H 1 (M; Z) = 0. Also, M has Lefschetz type trivially since H 1 (M; Z) = 0. However, we plainly see that we have 1

dim Q H ∗ (MS ; Q) = 4 < 8 = dim Q H ∗ (M; Q).

313

314

7 : G-spaces

Therefore, by Theorem 7.36, the Borel fibration is not TNCZ and the csymplectic analogue of Theorem 7.70 does not hold.

7.6.5

Hamiltonian bundles

If a T k action on (M2n , ω) is Hamiltonian, then the Borel fibration M → MT k → BT k is a bundle whose structure group (i.e., the torus T k ) lies inside the Hamiltonian diffeomorphisms of M. Recall that a Hamiltonian diffeomorphism of a symplectic manifold (M2n , ω) is a symplectomorphism φ : M → M which is the time-1 map of a flow generated by a family of Hamiltonian vector fields. This means that there are vector fields Xt with i(Xt )ω = dHt for all t and dφt = Xt ◦ φt . dt The Hamiltonian diffeomorphism is then given by φ = φ1 . The Hamiltonian diffeomorphisms form a subgroup of the symplectomorphism group of all diffeomorphisms preserving the symplectic form. For more information, see [188]. The situation of a Hamiltonian action can then be generalized as follows. Definition 7.72 A symplectic bundle F → E → B is Hamiltonian if (F, ωF ) is symplectic and the structure group of the bundle acts by Hamiltonian diffeomorphisms. The fundamental properties of Hamiltonian bundles may be found in [162]. In particular, the following is shown there. Lemma 7.73 ([162, Lemma 2.3]) If (F 2n , ωF ) → E → B is a symplectic bundle and B is simply connected, then the bundle is Hamiltonian if and only if there exists some cohomology class α ∈ H 2 (E; R) which restricts to the class [ωF ] ∈ H 2 (F; R) by i∗ (α) = [ωF ]. Note the similarity to Thurston’s condition in Theorem 4.91. In the lemma, however, we do not require B to be symplectic, but we already know that the bundle is symplectic. Of course, Thurston’s condition is just a sufficient condition, so the assumption that the bundle is symplectic does not automatically imply the existence of a class α. Lalonde and McDuff apply Lemma 7.73 to prove analogues of Theorem 7.70 for various types of Hamiltonian bundles. For instance, they prove that every Hamiltonian bundle over a product of complex projective spaces is TNCZ. Indeed, they ask the Question 7.74 (Lalonde–McDuff question) Is every Hamiltonian bundle TNCZ?

7.6


Here we want to consider the special situation where the fiber is a nilmanifold. We have already seen how to build models of nilmanifolds in Section 3.2 and, in particular, we know that all generators, {xj }, are in degree 1. Hence, when we build the model of a symplectic nilmanifold, the symplectic form is represented by an element ω = cjk xj xk made up of products of degree 1 elements. We can now give a result due to Stepien [243] that answers a special case of the Lalonde–McDuff question using models. Theorem 7.75 Let (M2n , ωM ) → E → B be a Hamiltonian bundle with B simply connected and (M2n , ω) a nilmanifold. Then the bundle is rationally trivial. In particular, the bundle is TNCZ. ρ

Proof Let (∧Y, dY ) → (∧Y ⊕ X, D) → (∧X, dX ) denote a relative Sullivan modelfor the bundle M → E → B. As discussed above, there is an element ω = cjk xj xk in (∧X, dX ) made up of products of generators xm ∈ X 1 whose cohomology class is [ωM ]. Note that, since ωn = 0 and the xj , j = 1, . . . , 2n are in degree 1, it must be the case that ω uses all of the generators in X 1 and ωn = C x1 x2 · · · x2n Because B is simply connected, Y 1 = 0, so the differential on the xj has the form Dxj = dX xj + yj , where yj ∈ Y 2 . Now, since the bundle is Hamiltonian, Lemma 7.73 says that there is some cohomology class mapping to the class of ω in H ∗ (∧X, dX ) under the projection ρ : (∧Y ⊕ X, D) → (∧X, dX ). If the cocycle τ represents this class, then ρ(τ ) − ω = dX γ , so in (∧Y ⊕ X, D) we have τ − b = ω + dX γ with b ∈ Y 2 (since Y 1 = 0). If we apply D, then we find 0 = D(ω + dX γ ), since every element of Y 2 is a cocycle and τ is a cocycle. Let ω˜ = ω + dX γ . So the Hamiltonian hypothesis has allowed us to change the representative of the symplectic cohomology class in M while keeping the same type of expression in terms of generators of (∧X, dX ); ω˜ = rjk xj xk . Now we can use the fact that ω˜ is a D-cocycle to infer that ω˜ n is also a D-cocycle. Hence, computing D ω˜ n = 0, we have 0=C (−1)j−1 D(xj )x1 x2 · · · xˆ j · · · x2n j

=C (−1)j−1 (dX xj + yj )x1 x2 · · · xˆ j · · · x2n j

=C (−1)j−1 yj x1 x2 · · · xˆ j · · · x2n j

(where xˆ j denotes deletion of xj ) since dX xj consists of products of the xk not including xj , the element x1 x2 · · · xˆ j · · · x2n contains all generators except for xj and elements in degree 1 have zero squares. But now, since

315

316

7 : G-spaces

{x1 x2 · · · xˆ j · · · x2n } is a set of linearly independent elements in ∧X, we see that we must have yj = 0 for j = 1, . . . , 2n. This means that Dxj = dX xj and, since X = X 1 , the relative Sullivan model is a tensor product of cdga’s. Hence, the bundle is rationally trivial. Exercises for Chapter 7 Exercise 7.1 Let M be a compact manifold with an almost free action of the compact Lie group G. We consider the projection q : EG ×G M → M/G , q[(x, m)] = [m]. Show that the fiber of q at x is (G/Gx )G and that (G/Gx )G ∼ = EG/Gx = BGx . Deduce now from the Leray spectral sequence, H p (M/G; H q (BGx ; Q)) ⇒ p+q HG (M; Q), an isomorphism ∗ (M; Q) . H ∗ (M/G; Q) ∼ = HG

Hint: see [144, page 37]. Exercise 7.2 Suppose Mi , i = 1, . . . , n, are even dimensional simply connected manifolds satisfying πeven (Mi ) ⊗ Q = 0 and dim πodd (Mi ) ⊗ Q ≥ 2. Show that every circle action on the connected sum M1 #M2 # · · · #Mn has a fixed point. (Hint: consider the Euler–Poincaré characteristic and the rational homotopy Lie algebra of the connected sum.) Exercise 7.3 Fix a map S3 × S3 → S6 of degree 1 and use it to pull back the unit sphere bundle of S6 to a S5 -bundle over S3 × S3 . Compute rk0 (X). Exercise 7.4 Let M be a simply connected compact manifold whose minimal model is (∧(x, y, z, t), d) with |x| = 3, |y| = 9, |z| = 6, |t| = 11, dx = dy = dz = 0, dt = xy − z2 . Compute rk0 (M). Exercise 7.5 Write n + 1 = (n1 + 1) + (n2 + 1) + (n3 + 1). Consider the action of S1 on CP(n) defined by t[z1 , . . . , zn+1 ] = [t 2 z1 , . . . , t 2 zn1 +1 , t 3 zn1 +2 , . . . , t 3 zn1 +n2 +2 , zn1 +z2 +3 , . . . , zn+1 ] . Show that the fixed point set is the union CP(n1 ) ( CP(n2 ) ( CP(n3 ). Exercise 7.6 Let G = S1 . Show that G can act on CP(4) with fixed point set the disjoint union of CP(3) and an isolated point, and that G can act on the quaternionic projective plane HP(2) with fixed point set S4 and an isolated point. Make the equivariant connected sum CP(4)#HP(2) by removing small invariant discs around the isolated fixed points and pasting together the boundaries. Show that you have an action of S1 on a space Z whose cohomology satisfies the hard Lefschetz property, and such that at least one component of the fixed point set does not satisfy the hard Lefschetz property (see [7]).

8

Blow-ups and Intersection Products

As we mentioned in the Preface, models play an important role in understanding various properties of complex and symplectic blow-ups. They also can be used to study features of the Chas–Sullivan loop product, a rather mysterious new topological tool. In this chapter we will consider two types of questions. First of all, let f : N → M be a closed submanifold of a compact manifold and denote by C its complement M\N. The natural problem is to know if the rational homotopy type of C is completely determined by the rational homotopy type of the embedding, and in that case to describe a model for the injection C → M from a model of the initial embedding. Now suppose that the normal bundle to the embedding has a complex structure. We then can of M along N, and ask if it is possible to determine take the blow-up M from the rational homotopy type of the the rational homotopy type of M embedding. P. Lambrechts and D. Stanley have given a positive answer to those questions. They prove that when dim M ≥ 2 dim N + 2, then the rational homotopy type of the complement C is completely determined by the minimal model of f . They also prove that when M is simply connected and depends only on dim M ≥ 2 dim N + 3, the rational homotopy type of M the rational homotopy type of the embedding and the Chern classes of the normal bundle [165]. We describe in Section 8.2 their construction of a model for M. We have seen in Chapter 4 that the assumption that a compact manifold is Kähler implies its formality. Since Kähler manifolds are the most basic examples of compact symplectic manifolds, a fundamental question is whether such a statement also holds for symplectic manifolds. In Section 8.2, we describe non-formal simply connected symplectic manifolds that are blow-ups. The key point here is that we can, in some explicit sense, actually understand the complete rational homotopy of the blow-up rather than just subsidiary properties (such as existence or nonexistence of Massey products). Such an understanding should prove important in the future.

318

8 : Blow-ups and Intersection Products

Our second question concerns the theory of cycles in a compact manifold. M. Chas and D. Sullivan extended the ideas behind standard intersection 1 theory to an intersection theory of cycles in LN = N S for any compact oriented manifold N [53]. They define a product, called the loop product, on H∗ (LN), Hp (LN) ⊗ Hq (LN) → Hp+q−n (LN) , that combines the intersection product on the chains on N and the Pontryagin composition of loops in N. Here, minimal models appear as a useful tool for making explicit computations of the loop product. As a corollary, we give a proof of the Cohen–Jones algebra isomorphism between H∗−n (LN; Q) and the Hochschild cohomology HH ∗ (C ∗ (N; Q), C ∗ (N; Q)) , where, as usual, C ∗ (N; Q) denotes the singular cochains on N with coefficients in Q.

8.1

The model of the complement of a submanifold

Let M be a compact orientable nilpotent m-dimensional manifold and suppose that f : N n → Mm is a smooth embedding of a compact nilpotent ndimensional submanifold. The image f (N) admits a tubular neighborhood which is a compact submanifold (with boundary) T ⊂ M of codimension 0 that deformation retracts to N. Let C be the closure of the complement of T in M, C = M\T. Then C and T are both compact manifolds with the same boundary ∂T. Denote by i : ∂T → T, j : C → M, g : ∂T → C, k : T → M and h : N → T the canonical injections. We then have a commutative diagram i

∂T g

C

j

/ T

k

/ M

Of course, C has the homotopy type of M\f (N) and the above diagram is a homotopy pushout. A key step is to understand the embedding of the complement algebraically. With this in mind, a model for j : C → M will be defined in Theorem 8.11 in terms of shriek maps and algebraic mapping cones.

8.1

8.1.1


Shriek maps

To an embedding, we can associate a cohomological shriek map defined as the composition s−m DN

H∗ (f )

DM

f ! : sn−m H ∗ (N; Q) −→ s−m H∗ (N; Q) −→ s−m H∗ (M; Q) ←− H ∗ (M; Q) , where s is the suspension functor (sk V)r = V r+k , and DN and DM denote the corresponding Poincaré duality isomorphisms. (Recall that we have the convention on homology that (H∗ )−p = Hp .) In particular, if [N] and [M] denote the homology orientation classes, then DN (α) = α ∩ [N] and f ! (sn−m v) ∩ [M] = H∗ (f )(v ∩ [N]) . By construction the cohomological shriek map f ! is a morphism of H ∗ (M; Q)-modules and an isomorphism in degree m. Cohomological shriek maps are based on the duality between cohomology and homology. We now extend this duality to models. Let (A, d) be a cdga whose cohomology H ∗ (A, d) satisfies Poincaré duality with a fundamental class ω ∈ Am . We denote by ε : (A, d) → (Q, 0) a map of differential A-modules of degree −m satisfying ε(ω) = 1. The map ε yields a duality map ψA : A → s−m A∨ = s−m Hom(A, Q) , ψA (1) = s−m ε , sm ψA (a)(b) = (−1)|b| ε(a · b) . (For this, recall that we have (s−m Hom(A, Q))r = (Hom(A, Q))r−m = Hom(Am−r , Q).) Lemma 8.1 ψA is a quasi-isomorphism of differential graded A-modules. Proof This is a reformulation of the fact that H ∗ (A) satisfies Poincaré duality. Let’s come back now to the smooth embedding f : N → M and let ϕ : (A, d) → (B, d) be a model for f . Definition 8.2 A shriek map for f associated to ϕ is a morphism of differential A-modules g : sn−m (B, d) → (A, d) that induces in cohomology the cohomological shriek map f ! : sn−m H ∗ (N; Q) → H ∗ (M; Q). From now on we will abuse notation and denote both the shriek map and the cohomological shriek map by f ! .

319

320


Example 8.3 Let ϕ : (A, d) → (A⊗∧W, d) denote a relative minimal model for f , and consider the diagram s−m+n (A ⊗ ∧W, d) ψA⊗∧W

(A, d)

s−m (A ⊗ ∧W, d)∨

ϕ∨

ψA

/ s−m (A, d)∨

where ψA and ψA⊗∧W are the quasi-isomorphisms defined in Lemma 8.1. Since (A ⊗ ∧W, d) is a semifree A-module, the lifting property of semifree models (Proposition 2.107) gives a map f ! : s−m+n (A ⊗ ∧W, d) → (A, d) making the diagram commutative up to homotopy. This map is a shriek map for f associated to ϕ. Lemma 8.4 The embedding f : N → M admits a model ϕ : (A, d) → (B, d) with the following properties: 1. A≥m+1 = 0 and B≥n+2 = 0; 2. there is a shriek map f ! : s−m+n (B, d) → (A, d) for f associated to ϕ. Proof We first define the cdga (A, d) as the quotient of the minimal model (∧V, d) of M by the ideal I = (∧V)>m ⊕C1 where C1 is a complement to the cocycles in (∧V)m . We then take a relative minimal model of f of the form (A ⊗ ∧W, d) and we obtain a shriek map g : s−m+n (A ⊗ ∧W, d) → (A, d) as in Example 8.3. We define J = (A ⊗ ∧W)≥n+2 ⊕ C2 , where C2 is a complement of the cocycles in (A ⊗ ∧W)n+1 . For degree reasons, g(J) = 0. We therefore define (B, d) = ((A ⊗ ∧W)/J, d). The maps f and g factor to give the required model and shriek map. Lemma 8.5 Let (A, d) be a finite type cdga whose cohomology satisfies Poincaré duality with top dimension m and let (M, d) be a finite type semifree A-module. Then the evaluation in cohomology induces an isomorphism ∼ =

[(M, d), (A, d)] −→ HomH(A) (H m (M, d), H m (A, d)) . Proof First, recall the functor Ext defined in Section 2.8. If (M, d) and (N, d) are differential graded A-modules, and (P, d) → (M, d) is a semifree model for (M, d), then p

ExtA (M, N) = H p (HomA ((P, d), (M, d))) , and [(M, d), (N, d)] = Ext0A (M, N) . There is a spectral sequence due to Moore [200] converging to p,q ExtA (M, N) whose E2 -term is ExtH(A) (H(M), H(N)), where the indices are

8.1


d

defined as follows. Let Ln → Ln−1 → · · · L0 → H(M) be a free resolution of H(M) as an H(A)-module. Then, since H(M) is a graded vector space, each Ln is also a graded vector space and d(Ln )q ⊂ (Ln−1 )q+1 . By definition, we then have p,q

p

ExtH(A) (H(M), H(N)) = ExtH(A) (H(M), H(N))p+q = (H p (HomH(A) ((L∗ , d), H(N))))p+q . In this notation p refers to the resolution degree of L∗ and p + q is the total degree. In our case (N, d) = (A, d) and we have p,q

ExtH(A) (H(M), H(A)) = (H p (HomH(A) ((L∗ , d), H(A))))p+q ∼ = (H p (HomH(A) (H(A)∨ , (L∗ , d)∨ )))p+q = (H p (HomH(A) (H(A), (L∗ , d)∨ )))p+q 0 if p > 0 p,q ∨ = H ((L∗ , d) ) = q H (M) if p = 0 . Here we have used the fact that when C and D are finite dimensional Rmodules, then taking the dual map gives an isomorphism HomR (C, D) → p,q HomR (D∨ , C ∨ ). Since E2 = 0 when p = 0, the spectral sequence collapses, and therefore, the morphism that gives the above induced map in cohomology, p

ExtA (M, A) → HomH(A) (H m−p (M), H m (A)) is an isomorphism. In particular [(M, d), (A, d)] ∼ = HomH(A) (H m (M), H m (A)) .

Corollary 8.6 Let ϕ : (∧V, d) → (∧W, d) be the minimal model of an embedding f : N → M and let θ : (P, d) → (∧V, d) be a morphism of (∧V, d)-modules which is an isomorphism on H m and such that (P, d) is a semifree model for s−m+n (∧W, d)∨ . Then there is a constant q such that q · θ is a shriek map for the embedding. This means that shriek maps are defined in a unique way by what they do on cohomology in degree m.

8.1.2

Algebraic mapping cones

An important tool in our construction of a model for the complement C is the mapping cone construction associated to a morphism of differential graded A-modules. Let’s recall that now.

321

322


If f : (X, dX ) → (Y, dY ) is a morphism of differential graded A-modules, then the algebraic mapping cone of f is the differential graded A-module i i C(f ) = Y ⊕f sX := ⊕∞ i=1 Y ⊕ (sX) ,

with differential given by d(y, sx) = (dY y + f (x), −sdX x). It is easy to see that d 2 = 0. Also, we can define the morphisms i : Y → Y ⊕f sX and τ : Y ⊕f sX → X by i(y) = (y, 0),

τ (y, sx) = x.

The mapping cone construction is invariant with respect to quasiisomorphisms as the next result demonstrates. Lemma 8.7 Suppose we have a commutative diagram of differential graded A-modules, f

(X, d) O

/ (Y, d) O

h g

(Z, d)

k

/ (T, d)

where h and k are quasi-isomorphisms. Then we have the following commutative diagram in which the horizontal maps are the natural injections and is also a quasi-isomorphism. (Y, d) O

/ C(f ) O

k

(T, d)

/ C(g)

In particular C(f ) and C(g) are quasi-isomorphic. Now, in Example 8.3 and Lemma 8.4, we gave different constructions of a shriek map. By Lemma 8.7, we see that, in fact, they have quasi-isomorphic mapping cones. g

Lemma 8.8 Let 0 → (X, d) → (Y, d) → (Z, d) → 0 be a short exact sequence of differential graded A-modules. Then there is a quasiisomorphism of A-modules C(g) → (Z, d) making the following diagram

8.1


commutative. (Y, d)

/ (Z, d) HH O HH HH HH H#

C(g) Proof Consider the diagram of short exact sequences 0

/ (X, d) O

g

/ (Y, d) O ϕ

=

0

/ (X, d)

/ (Z, d) O ϕ¯

/ (X ⊕ Y ⊕ sX, D ) O

/ C(g) O

=

(Y, d)

/ C(g)

θ

/ 0

/ 0

Here D (x) = dx , D (y) = dy , D (sx) = g(x)−sdx−x, ϕ(x) = g(x) , ϕ(y) = y , ϕ(sx) = 0 and θ(y) = y. Since ϕ is a quasi-isomorphism, the same is true for ϕ. ¯ Interesting examples of short exact sequence of differential modules arise over the real numbers from the inclusion of relative de Rham forms associated to the injection j : C → M: ADR (j)

0 → ADR (M, C) → ADR (M) −→ ADR (C) → 0 . Over the rational numbers, the situation is similar. We have that the morphism APL (j) : APL (M) → APL (C) is surjective with kernel denoted by APL (M, C) and this then gives the short exact sequence of APL (M)-modules, APL (j)

0 → APL (M, C) → APL (M) −→ APL (C) → 0 . Lemma 8.9 If f , g : (X, d) → (Y, d) are two homotopic maps of differential graded A-modules, then the algebraic mapping cones C(f ) and C(g) are quasi-isomorphic differential A-modules and we have a commutative diagram (Y, d)

/ C(f ) HH HH HH θ HH H#

C(g)

323

324


Proof By hypothesis (see Definition 2.106), there is a map of A-modules of degree −1, H : X → Y such that f − g = dH + Hd. We define a map θ : C(f ) → C(g) by putting θ(y) = y and θ(sx) = sx + Hx. The morphism θ commutes with the differentials, makes the diagram above commutative and is a quasi-isomorphism of differential A-modules.

8.1.3

The model for the complement C

In the previous section, we worked with models as differential modules. Here, we add hypotheses to obtain models in the framework of cdga’s. Suppose that we have a smooth embedding f : N n → Mm and ϕ : (A, d) → (B, d) is a model for f with associated shriek map f ! : sn−m (B, d) → (A, d). Since H ≥m (C(f ! )) = 0, C(f ! ) contains an acyclic submodule I such that C(f ! )≥m ⊂ I. When m ≥ 2n+2, the quotient C(f ! )/I becomes a differential graded algebra with multiplication defined by sx · sy = 0 ,

v · sy = (−1)|v| s(v · y) ,

x, y ∈ sn−m B , v ∈ A .

Theorem 8.10 ([167]) Suppose Mm and N n as above are simply connected. If m ≥ 2n + 1, then the induced morphism ϕ : (A, d) → C(f ! )/I , ϕ(x) = x is a cdga model for the embedding j : C → M. More generally, we have Theorem 8.11 ([169]) Suppose there is an integer r ≥ 0 such that Hq (f ; Q) is an isomorphism for q ≤ r and suppose that m ≥ 2n + 2 − r. Let I be an acyclic submodule in C(f ! ) containing C(f ! )≥m−r . Then the complex C(f ! )/I is a differential graded algebra and the induced morphism (A, d) → C(f ! )/I is a cdga model for the embedding j : C → M. As a consequence, under the dimension hypothesis of the theorem, the rational homotopy type of the complement C depends only on the rational homotopy type of f . Proof Note that, if Hq (f ; Q) is an isomorphism for q ≤ r, then by Poincaré duality, H q (f ! ) is an isomorphism for q ≥ m−r. In particular the submodule I of C(f ! ) defined by I = S ⊕ C(f ! )≥m−r , where S is a complement of the cocycles in degree m − r − 1, is acyclic. We use the notation at the beginning of Section 8.1. Denote by µ : (∧V, d) → APL (M) a minimal model for M, and by ϕ : (∧V, d) →

8.1


(∧V ⊗ ∧W, d) a relative minimal model for k, APL (k)

APL (M) O µ

/ APL (T) O

APL (h)

/ APL (N)

ϕ

(∧V, d)

/ (∧V ⊗ ∧W, d)

By composition with the quasi-isomorphism APL (h), ϕ can be seen to be a relative minimal model for f , with associated shriek map f ! : sn−m (∧V ⊗ ∧W, d) → (∧V, d). Observe first that, by the Thom isomorphism, the multiplication by the Thom class induces a quasi-isomorphism of APL (M)-modules

sn−m APL (T) → APL (T, ∂T) . Now, by Mayer–Vietoris, the restriction of forms APL (M, C) → APL (T, ∂T) is also a quasi-isomorphism of APL (M)-modules. By the lifting homotopy property of semifree models (Proposition 2.107 (3)) applied to the diagram

/ APL (T, ∂T) o APL (M, C) 3 g g g g g g g g gg g g g

sn−m APL (T) O

APL (M)

sn−m (∧V ⊗ ∧W, d)

we get a quasi-isomorphism of (∧V, d)-modules g : sn−m (∧V ⊗ ∧W, d) → APL (M, C) . Denote by ρ : APL (M, C) → APL (M) the canonical injection. Then, if we write (P, d) = sn−m (∧V ⊗ ∧W, d), we have two maps of (∧V, d)-modules from (P, d) to APL (M): the morphisms µ ◦ f ! and ρ ◦ g. (P, d)

(P, d)

f!

g

/ (∧V, d)

/ APL (M, C)

µ

ρ

/ APL (M)

/ APL (M)

Both maps induce isomorphisms in cohomology in degree m, so they differ by multiplication by a rational number q. By Lemma 8.5, this implies that µ ◦ f ! is homotopic to ρ ◦ (qg).

325

326


Using Lemmas 8.7, 8.8 and 8.9,we have a sequence of quasiisomorphisms of differential (∧V, d)-modules. / (∧V ⊕f ! sP, D)

(∧V, d) µ

APL (M)

/ (APL (M) ⊕µ◦f ! sP, D)

=

APL (M)

/ (APL (M) ⊕ρ◦(qg) sP, D)

=

APL (M)

/ APL (M) ⊕ρ sAPL (M, C)

=

APL (M)

APL (j)

/ APL (C)

Now let (∧V, d) → (∧V ⊗ ∧Z, D) be a relative minimal model for the embedding j : C → M. Since (∧V ⊗ ∧Z, D) is a semifree (∧V, d)-module, we have a commutative diagram (∧V, d)

/ (∧V ⊗ ∧Z, D) NNN NNN NNN θ NN'

(∧V ⊕f ! sP, D)

where θ is a quasi-isomorphism of (∧V, d)-modules. Since (sP)n = 0 and Bn = Q uN (where uM and uN are the respective fundamental classes). A model for f is given by the trivial map ϕ : (A, d) → (B, d). This means that the A-module structure on B is trivial: A+ · B = 0. A shriek map is then given by f ! : sn−m (B, d) → (A, d) ,

f ! (sn−m (uN )) = uM ,

f ! (sn−m Bn = 0, and there are graded bases ai , a i such that (ai · a j ) = δij ω for all i and j. Here, the duality map ψA : (A, d) → s−n (A, d)∨ defined in Lemma 8.1 is an isomorphism of differential graded A-modules (rather than simply being only a quasi-isomorphism). Definition 8.16 The element DA = i (−1)|ai | ai ⊗ a i ∈ A ⊗ A is called the diagonal cocycle of A. The diagonal DA is a cocycle, satisfies the useful equality (a ⊗ 1) · DA = (1 ⊗ a) · DA , and induces in cohomology the usual diagonal element DH ∗ (A) defined in the analogous way on H ∗ (A, d).

8.1


This diagonal cocycle has a very geometric meaning. Consider the diagonal injection : N → N × N of a 2-connected compact manifold N. The complement C is the configuration space of two points in N, and we know how to construct a model of this complement from a shriek map associated to . The multiplication map (A, d) ⊗ (A, d) → (A, d) makes (A, d) into a differential graded (A ⊗ A)-module. Moreover (see Example 2.48), by the property above, the multiplication by DA , µDA : s−n A −→ A ⊗ A ,

s−n a → DA · (1 ⊗ a)

is a morphism of differential graded (A ⊗ A)-modules. The next result gives the link between the shriek map and the map µDA . Lemma 8.17 The multiplication by DA , µDA is a shriek map for the diagonal injection : N → N × N. Proof Recall that the multiplication m : A ⊗ A → A is a model for the embedding . The lemma follows now from the commutativity of the following diagram in which the vertical maps are isomorphisms. s−n A s−n ψA

s−2n A∨

µDA

m∨

/ A⊗A

ψA ⊗ψA

/ s−n A∨ ⊗ s−n A∨

8.1.5

The configuration space of two points in a manifold

Let N be a 2-connected n-dimensional compact manifold and let : N → N × N be the diagonal submanifold. The purpose of this subsection is to give a description of a model for the complement of the diagonal: that is, the configuration space of two points in N: F(N, 2) = {(x, y) ∈ N × N | x = y}. Theorem 8.18 ([166]) Let N be a compact 2-connected manifold, and suppose (A, d) is a Poincaré duality model for N with associated diagonal cocycle DA . Then the quotient map A ⊗ A → (A ⊗ A)/(DA ), where (DA ) is the ideal generated by DA , is a model for the embedding j : F(N, 2) → N × N. Proof We use the notation of Subsection 8.1.4. Since µDA is a shriek map for , a model for F(N, 2) as an A⊗A-module is given by the mapping cone

329

330


C(µDA ). Since N is 2-connected, H≤2 (j; Q) is an isomorphism. Therefore, by Theorem 8.11, there is an acyclic submodule I containing C(µDA )≥m−2 such that the quotient map A ⊗ A → C(µDA )/I is a model for the injection F(N, 2) → N × N. Note that DA = 1 ⊗ ω + ... . This implies that the A-submodule generated by DA is a free module of rank one. In particular, the multiplication by DA defines an isomorphism between s−n A and the ideal generated by DA . Therefore the quotient map q : C(µDA )/I → (A⊗A)/(DA ) is a quasi-isomorphism of differential graded algebras. It follows that the other quotient map A ⊗ A → (A ⊗ A)/(DA ) is a model for the injection j : F(N, 2) → N × N. The homotopy type of the configuration space of k points in N will be described more generally in Section 9.1.

8.2

Symplectic blow-ups

As symplectic topology became an active area in the mid 1980s and 1990s, certain fundamental questions naturally arose. For instance, since a compact Kähler manifold is always symplectic and since the standard symplectic examples are Kähler manifolds (e.g. CP(n)), it was wondered whether all compact symplectic manifolds were Kähler. We have seen that compact Kähler manifolds satisfy the hard Lefschetz property (see Theorem 4.35) and it was shown in Theorem 4.98 that a nilmanifold can only have Lefschetz type if it is a torus. By Proposition 4.94, a nilmanifold is symplectic when it has a degree 2 cohomology class which multiplies up to a top class, so it is therefore fairly easy to find nonsimply connected compact symplectic manifolds that are not Kähler (see [257] for more information on this issue). However, the simply connected case took more work and it wasn’t until 1984 that McDuff [188] used the symplectic blow-up construction (modeled after the blow-up in complex geometry) to construct a compact symplectic manifold that did not satisfy the hard Lefschetz Property and, therefore, could not be Kähler (with respect to any metric). So finer questions then arose. It was shown in [71] (see Theorem 4.43) that compact Kähler manifolds are formal spaces. In particular, of course, this entails the vanishing of all Massey products. Even though not every symplectic manifold is Kähler, could every compact symplectic manifold be formal? Again, the non-simply connected case is easy by Proposition 3.20 and the simply connected case is harder. In [257] it was implied that blow-ups were the place to look

8.2

Symplectic blow-ups

for counterexamples to formality for symplectic manifolds in the simply connected case and, indeed, Babenko and Taimanov soon thereafter demonstrated that a certain blow-up has Massey products, so could not be formal (see [17]). (This approach was later generalized in [230].) The blow-up construction is fundamental in symplectic (and complex) topology, so if rational homotopy methods are to find real use there, it is essential to know more information about blow-ups than just the existence or nonexistence of nontrivial Massey products. Namely, we want to know the rational homotopy type of blow-ups and, for this, the construction of an algebraic model is required. In this section, we shall give background on the blow-up construction and in the next section, we shall describe the model.

8.2.1

Complex blow-ups

The complex blow-up of a complex surface (or rather the blow-down) originated as a way to remove certain negative intersections in homology obstructing possible embeddings in some CP(n). See [234] for a brief exposition along these lines. The idea for a complex surface W and a point p ∈ W is the following. Take a neighborhood T of p given by a holomorphic chart and a disk about (0, 0) ∈ C2 and, using those identifications, take the bundle over CP(1), LT = {(t, l) ∈ T × CP(1) | t ∈ l}, and note that T\p is bi-holomorphic to LT \{p× CP(1)} since t = p uniquely determines a complex line through the origin in C2 . This means that the neighborhood T can be replaced by LT and, effectively, p has been “blown up” to a CP(1). This CP(1) is usually denoted by E and is called the excep7 It has the property that its tional curve in the blow-up of W at p, W. self-intersection is E · E = −1. Now, again thinking of T as a complex 2-disk centered at p, we see that any complex line l intersects T in a real 2-disk Dl . Hence, T is a union of real 2-disks which intersect only at p. Of course, the intersections of the disks with ∂T are the orbit circles on ∂T = S3 of the Hopf action. The corresponding disks in LT are simply the Dl , but now taken to be disjoint. Indeed, the centers of the disks correspond to the lines l, so the centers form a CP(1). So we see that LT is the (total space of the) disk bundle over CP(1) whose boundary is the associated Hopf sphere bundle 7 cut out T (with ∂T = S3 recall) and S3 → CP(1). To form the blow-up W, 3 glue in LT (also with boundary S ). Furthermore, it is well-known that LT is the normal disk bundle to the embedding CP(1) → CP(2), so we may think of LT as a tubular neighborhood of CP(1) in CP(2). Of course, since CP(2) = CP(1) ∪ e4 , for a 4-cell e4 attached by the Hopf map, if we delete

331

332


an open 4-disk D4 disjoint from CP(1), we have CP(2) − D4 LT . Then, by attaching LT to W, we should obtain a connected sum. This heuristic argument can be made precise (see [234, section 7.1], for instance) to obtain Proposition 8.19 7 is diffeomorphic to the connected sum 1. The complex surface W 7 = W#CP(2). W 72n obtained by blowing 2. In higher dimensions, the complex manifold W up a point is diffeomorphic to the connected sum 7 = W#CP(n). W Such a connected sum is often called the topological blow-up of W at a point. The bar over CP(2) (and CP(n)) denotes CP(2) (and CP(n)) with its orientation reversed. This is a way to obtain an oriented manifold whose orientation is compatible with the orientations of the summands. In symplectic geometry, the role of the blow-up point is taken by a symplectically embedded ball in order to describe a new symplectic form. Since a ball is contractible, the symplectic and complex constructions give the same diffeomorphism type when they both can be defined. See [189, page 239–251] for details.

8.2.2

Blowing up along a submanifold

The blow-up construction can be extended to submanifolds and this is now what we will focus on in the symplectic context. Suppose f : N n → Mm is a codimension 2k submanifold whose normal bundle ν has a complex structure. If (N, ω|N ) ⊂ (M, ω) is a symplectic submanifold, then this hypothesis always holds for the following reasons. First, the nondegeneracy of ω and of ω|N leads to a splitting TM = TN ⊕ TN ω , where TN ω denotes the ω-complement of TN in TM. Of course, we have the usual isomorphism ν∼ = TM/TN, where ν is the normal bundle of the embedding, so we obtain ν ∼ = TN ω . Now, ω|TN ω is a nondegenerate skew-symmetric bilinear (i.e. symplectic) form on each fiber, thus reducing the structure group of ν to Sp(k, R) U(k) (see Exercise 1.3) and making ν a complex bundle with fiber Ck . Denote the unit disk bundle of ν by Dν and the associated sphere bundle by Sν. As above, let T denote a tubular neighborhood of N in M diffeomorphic to Dν with boundary ∂T diffeomorphic to the sphere bundle Sν. Also, denote by C the closure of the complement of T in M, C = M\T. The complex structure on ν implies that the circle S1 (thought of as S1 ⊂ C∗ ) acts on the sphere bundle S2k−1 → Sν = ∂T → N to give the projectivized bundle CP(k − 1) → Pν → N. Note that there is a bundle map ∂T → Pν.

8.3

Symplectic blow-ups

of M along N is the smooth manifold Definition 8.20 The blow-up M = C ∪∂T Pν M obtained from the pushout diagram ∂T q

Pν

k

/ C / M

Remark 8.21 The definition above gives the homotopy type of the blow-up because the map k : ∂T → C is a cofibration and this makes the pushout a homotopy pushout as well. In particular, this means that any of the constituent spaces may be replaced by spaces of the same homotopy type To obtain the diffeomorphism without changing the homotopy type of M. type of the blow-up requires a bit more work and a larger space than Pν (which has the same homotopy type). This then allows the definition of the → M. See [257, Chapter 4] for instance. blow-up projection φ : M The algebraic topology of the blow-up may be understood in terms of the cohomologies of the spaces involved in its construction. In particular, we have the following fundamental result. Recall first that if ci ∈ H 2i (N; Z) denote the Chern classes of the bundle ν, then the cohomology of Pν is given by H ∗ (Pν; Z) ∼ = H ∗ (V)[a]/ak + c1 ak−1 + . . . + ck where a ∈ H 2 (Pν; Z). By Van Kampen and Mayer–Vietoris, we have Theorem 8.22 With the notation above: ∼ • π1 (M) = π1 (M). • There is a short exact sequence

Z) → A∗ → 0 0 → H ∗ (M; Z) → H ∗ (M; where the quotient A∗ is the ideal generated by a in H ∗ (Pν; Z): A∗ = a · H ∗ (N; Z)[a]/ak + c1 ak−1 + . . . + ck . Once these algebraic properties of the blow-up have been given, it still has a symplectic structure. This can be done, but remains to show that M we omit details and simply state the result. of (M, ω) along Theorem 8.23 ([188]) If M is compact, then the blow-up M (N, ω|N ) has a symplectic form ω.

333

334


8.3

A model for a symplectic blow-up

Throughout this section we assume that M is simply connected and that dim M ≥ 2 dim N + 3. This latter condition is only required to guarantee that any model of the embedding f : N n → Mm may be used for the construction of the blow-up model. (Indeed, without this condition, homotopic embeddings with isomorphic normal bundles can be defined that give non-rationally equivalent blow-ups [168].)

8.3.1

The basic pullback diagram of PL-forms

The key to obtaining a model for the blow-up is the fundamental pushout diagram ∂T

k

/ C

q

/ M

Pν

A model for the pushout is given by the pullback of the corresponding homomorphisms of Sullivan PL-forms (see Section 2.4). The pushout diagram induces a commutative diagram / APL (C)

APL (M)

APL (Pν)

APL (q)

APL (k)

/ APL (∂T)

Note that, since k : ∂T → C is the inclusion of a sub-complex, APL (k) is surjective. Therefore, the universal property for pullbacks provides a cdga homomorphism → APL (C) ×A (∂T) APL (Pν) φ : APL (M) PL and we have the following. Proposition 8.24 (see [87, Proposition 13.5]) φ is a quasi-isomorphism.

8.3.2

An illustrative example

Example 8.25 Let MPν → M∂T ← MC denote the induced morphisms of minimal models. It would be very nice if a model for the blow-up could be obtained as a pullback of these morphisms. Unfortunately, the only time

8.3


we can guarantee this to be the case is when at least one of the morphisms is surjective (see [87, Proposition 13.6]). Of course, there is one natural geometric situation where surjectivity holds. Namely, consider blowing up a point p in a complex manifold M2n (where 2n is the real dimension). The neighborhood T (as we saw above) is a 2n-ball with ∂T = S2n−1 and trivial normal bundle ν. Therefore, Pν CP(n − 1) and the map ∂T → Pν is simply the Hopf bundle map S2n−1 → CP(n − 1). The minimal models are, respectively, (∧(z), d = 0) and (∧(x, y), dy = xn ), where the degrees of generators are |z| = 2n − 1, |x| = 2 and |y| = 2n − 1. The long exact homotopy sequence of the bundle S1 → S2n−1 → CP(n − 1) shows that Z = π2n−1 (S2n−1 ) ∼ = π2n−1 (CP(n − 1)), so the identification of (the duals of) the rational homotopy groups with the sets of generators (see Theorem 2.50) then implies that y maps to z under the induced model morphism (∧(x, y), dy = xn ) → (∧z, d = 0). Since we are over Q and (∧z, d = 0) is an exterior algebra, this means that the morphism is surjective. By [87, Propositions 13.3, 13.6], we then obtain a model of the blow-up (∧(x, y), dy = xn ) ×∧z MC , where C = W − T = W − D4 and MC is a model for C [87, Proof of Theorem 38.5]. Let (A, d) be a Poincaré duality model for M with fundamental class uA . A model for the embedding C → M is thus given by the injection (A, d) → (A ⊕ Q · u, d) by letting du = uA and setting u · A+ = 0. Moreover, the model of the injection S2n−1 → C is given by the morphism (A ⊕ Q · u, d) → (∧(z), d = 0) that maps u to z. Together with y → z, we have the pullback of the models, (∧(x, y), dy = xn ) ×∧z (A ⊕ Q · u, d) = (∧(x, y) ⊕Q A , dy = xn + ua ) . This model is a model for the connected sum W#CP(n) – just as it should be by Proposition 8.19 (2). We will also see this when we describe the general model for a blow-up. Indeed, this example contains many of the ingredients (at least in spirit) of the general construction of the model for a blow-up.

8.3.3

The model for the blow-up

Let’s first fix notation. The morphism φ : (A, d) → (B, d) is a model for the embedding f : N n → Mm with m − n = 2k. We suppose (without loss of generality by Lemma 8.4) that A≥m+1 = 0, B≥n+1 = 0 and there is a shriek map f ! : sn−m (B, d) → (A, d). We let (P, d) = sn−m (B, d), and denote by γi ∈ (B, d) representatives for the Chern classes ci (ν) of the normal bundle ν (with γ0 = 1 and γk = 0 since 2k > dim V by assumption).

335

336


Proposition 8.26 1. A model for q : ∂T → Pν is given by the projection ¯ proj : (B ⊗ ∧(x, z), D) → (B ⊗ ∧(z), D), x → 0 k−i , Dz ¯ = 0, and where |x| = 2, |z| = 2k − 1, Dx = 0, Dz = k−1 i=0 γi x 2k = m − n. 2. A model for k : ∂T → C is given by the homomorphism of algebraic mapping cones φ ⊕ id : A ⊕f ! sP → B ⊕φ◦f ! sP . Proof Since H ∗ (Pν; Q) = H ∗ (N; Q)[a]/ for the bundle Pν → N,

k−1 i=0

γi xk−i , in the relative model

(B, d) → (B ⊗ ∧(x, z), D) → (∧(x, z), D) , k−i . We now have only to we necessarily have Dx = 0 and Dz = k−1 i=0 γi x recall that q : ∂T = Sν → Pν is a morphism of bundles over N that maps π2k−1 (CP(k − 1)) isomorphically onto π2k−1 (S2k−1 ). By our dimension hypothesis, the second part of the proposition follows directly from Proposition 8.12. Note that, by our connectivity hypothesis, the two models for APL (∂T) 7 → APL (C) ×A (∂T) APL (Pν) is a quasiare isomorphic. Since φ : APL (W) PL isomorphism, we have the following definition and result. Definition 8.27 (Description of the model for the blow-up) Assume that m ≥ 2n + 3. Then, with the notation above, define

B l(A, B) = (A ⊕ (B ⊗ ∧+ (x, z)), D) with |x| = 2 and |z| = m−n−1 = 2k−1. The algebra structure on B l(A, B) is induced by the multiplications on A and B ⊗ ∧+ (x, z) and the formula a · (b ⊗ w) = (φ(a) · b) ⊗ w. The differential D on B l(A, B) is defined by D(a) = dA a D(b ⊗ x) = dB b ⊗ x

⎛

D(b ⊗ z) = dB b ⊗ z + (−1)|b| ⎝φ ! (s−2k b) +

k−1

⎞

(b · γi ) ⊗ xk−i ⎠ ,

i=0

where the γi ∈ B are representatives for the Chern classes, ci (ν), of the normal bundle ν.

8.3


Theorem 8.28 ([167]) With the above notation, when m ≥ 2n + 3, then and the injection A → B l(A, B) is a model for B l(A, B) is a model for M the blow-up projection M → M.

8.3.4

McDuff’s example

A theorem of Tischler (see [251]) says that a closed symplectic manifold (V 2m , ω) with integral symplectic form (i.e. [ω] ∈ H 2 (V; Z)) may be embedded symplectically in some (CP(n), α) for n large enough. So now consider such an embedding for the Kodaira–Thurston manifold, 7P(n). In order to satf : (KT, ω) → (CP(n), α) and take the blow-up C isfy our usual hypothesis dim M ≥ 2 dim N + 3, we take n ≥ 6, but in fact, simply according to Tischler, we could take n = 5. Let’s now find the model 7P(n). B l(A, B) of C We start by taking the model (A, d) = (∧(a)/(an+1 ), d = 0) for CP(n) with a → α. For V = KT, we take the model from Example 4.95, (B, d) = (∧(u, v, y, t), du = 0, dy = 0, dv = uy, dt = 0). Observe that the nonformality of KT is expressed by the Massey product vy (see Proposition 2.90 and Proposition 3.20). Note that these models satisfy the dimension restrictions imposed by Proposition 8.26. The form of B l(A, B) is then

B l(A, B) = (∧(a)/(an+1 ) ⊕ ∧(u, v, y, t) ⊗ ∧+ (x, z), D), with |z| = (2n − 4) − 1 = 2n − 5. We must now define the differential D and for that we need the shriek map and the Chern classes of the normal bundle. By Theorem 8.32, we know that the total Chern class of KT is trivial, c(KT) = 1. The total Chern class of CP(n) is c(CP(n)) = (1 + α)n+1 , so the Whitney product formula gives c(ν) = c(KT)c(ν) = f ∗ c(CP(n)) = f ∗ ((1 + α)n+1 = 1 + (n + 1)ω +

n(n + 1) 2 ω 2

since f ∗ (α) = ω and dim KT = 4. A general fact that we will not prove (see [165, Lemma 8.3]) is that the shriek map for any embedding f : V 2m → CP(n) is (for 2r = 2n − 2m), f ! : s−2r B → A,

f ! (ωj ) = V aj+r

where a → α under the model map and V is the coefficient relating ωm and the orientation uV given by the almost complex structure: ωm = V uV . For V = KT, by Proposition 4.94 we have a symplectic class ω = uv + yt with

337

338


ω2 = 2 uvyt, so KT = 2. Therefore we have s−2r (1) = 2ar = 2an−2 and s−2r (ω) = 2a1+r = 2an−1 . Since ω = uv + yt, we can define f ! (uv) = an−1 and f ! (yt) = an−1 . Finally, we have s−2r (ω2 ) = 2a2+r = s−2r (2uvyt) = 2s−2r (uvyt) which says s−2r (uKT ) = a2+n−2 = an = uCP(n) and f ! (s−2r ξ ) = 0 for any other monomial. Here this translates into a differential with D(a) = 0 D(q ⊗ x) = dB q ⊗ x D(q ⊗ z) = dB q ⊗ z + (−1)

|q|

f ! (s−2r q) + q ⊗ xn−2

n−3

+q(n + 1)ω ⊗ x

n(n + 1) 2 n−4 ω ⊗x , +q 2

using the fact that γj = 0 for j > 2 since dim KT = 4. Similarly, the term q (n(n + 1)/2) ω2 ⊗ xn−4 = 0 unless q = 1 for degree reasons. This description of D has the following consequences. D(v ⊗ x2 ) = D((v ⊗ x)(1 ⊗ x)) = D(v ⊗ x) · (1 ⊗ x) + (−1)3 (v ⊗ x) · D(1 ⊗ x) = (dB v ⊗ x) · (1 ⊗ x) = (uy ⊗ x) · (1 ⊗ x) = (u ⊗ x) · (y ⊗ x). This says that [u ⊗ x] · [y ⊗ x] = 0. Of course, we also have [y ⊗ x] · [y ⊗ x] = 0 since y has odd degree, so the Massey product (see Definition 2.89) [u ⊗ x], [y ⊗ x], [y ⊗ x] is defined with representative (v ⊗ x2 ) · (y ⊗ x) − (−1)3 (u ⊗ x) · 0 = (v ⊗ x2 ) · (y ⊗ x) = (vy ⊗ x3 ). Since v ⊗ x2 is not a cocycle (and from the form of the differential D if |z| = 7), we see that [vy ⊗ x3 ] is not in the ideal generated by [u ⊗ x] and [y ⊗ x]. Hence, [u ⊗ x], [y ⊗ x], [y ⊗ x] is a nontrivial Massey product. Therefore, by Proposition 2.90, we see that the blow-up of CP(n) along 7P(n), is not formal. Thus we have the following result. KT, C Theorem 8.29 ([17]) There exist closed simply connected symplectic manifolds that are not formal. Of course, the description given above using the model B l(A, B) shows exactly how Massey products can propagate from nonformal submanifolds to blow-ups along them. For a general discussion of this, see [230]. Theorem 8.29 powerfully illustrates the gap between Kähler manifolds and symplectic manifolds, even in the simply connected case.

8.3


Remark 8.30 In [188], D. McDuff constructed a blow-up of CP(5) along an embedding of the Kodaira–Thurston manifold KT → CP(5). She proved that if the hard Lefschetz property failed for M, then it also failed for the blow-up (see [188, Proposition 2.5] or [257, Chapter 4] for details). G. Cavalcanti [51] has shown that we can’t decide whether blow-ups have or don’t have the hard Lefschetz property based simply on the ambient space. He also shows that there are compact simply connected symplectic blow-ups which satisfy hard Lefschetz, but which are nonformal. On the other hand, there are compact simply connected symplectic blow-ups (e.g. McDuff’s example) which do not satisfy hard Lefschetz, so, by Theorem 4.82, they are simply connected counterexamples to the Brylinski conjecture.

8.3.5

Effect of the symplectic form on the blow-up

How much of an effect does the choice of symplectic form have on the rational homotopy type (and hence the diffeomorphism type) of a blow-up? In [165], embeddings are taken according to the Tischler theorem, f

CP(1) → CP(5),

f ∗ (α5 ) = α1 ,

where αi is the standard Kähler form on CP(i), i = 1, 5 and ∈ Z. Then, 7P (5) and using the model B l(A, B) denoting the resulting blow-ups by C (and, in particular, the multiplication derived from the R-dgmodule structure), the following result is shown. Theorem 8.31 ([165, Section 8.5]) If 1 = 2 , then the rational homotopy 7P (5) are different. Hence, there are an infinite 7P (5) and C types of C 1 2 number of rationally distinct blow-ups of CP(5) along CP(1) corresponding to the infinite number of integral symplectic forms on CP(1), α1 for ∈ Z.

8.3.6

Vanishing of Chern classes for KT

A key point in the construction of McDuff’s example in Subsection 8.3.4 was the triviality of the Kodaira–Thurston manifold’s total Chern class. Here we give the details of this vanishing result since it does not seem to be something that is widely known. The proof relies on an old construction that goes back at least to Borel and Hirzebruch’s work on the characteristic classes of homogeneous spaces [34]. This construction is known as the tangent bundle along the fibers. i

p

Suppose F → E → B is a fiber bundle of smooth manifolds with associated principal bundle G → P → B, where G right-acts smoothly on P and G left-acts smoothly on F. Then G left-acts smoothly on the tangent bundle TF of F by derivative maps. We can then form the quotient θ = P ×G TF

339

340


and, since E = P ×G F, we obtain the following map of bundles TF F

τ

i

/ θ / E

The bundle θ is called the (tangent) bundle along the fibers. The essential properties of the bundle along the fibers are delineated in [34]: 1. The tangent bundle TF is a pullback of θ: i∗ (θ ) = TF. 2. If G preserves some other structure on TF, then θ inherits the same structure and τ is a bundle map preserving that structure. In particular, if F is almost complex and G preserves the almost complex structure, then τ is a bundle map of almost complex structures. By (1), we have i∗ (c(θ )) = c(TF), where c(−) denotes the total Chern class. p

3. If F → E → B is a smooth bundle of manifolds, then the tangent bundle of E is given by TE ∼ = θ ⊕ p∗ (TB) (see [34, Proposition 7.6]). Our goal is to prove the following result (also see Exercise 8.3). Theorem 8.32 The Chern classes of the Kodaira–Thurston manifold vanish. That is, c1 (KT) = 0 and c2 (KT) = 0. Proof First, we note that all nilmanifolds have free circle actions on them, so their Euler characteristics vanish. For KT, this also follows by the product equality for the Euler characteristic of a fibration. Because c2 (KT) is the Euler class, we see that c2 (KT) = 0. Hence, we focus on the vanishing of c1 (KT). We now take the bundle along the fibers of the Borel fibration associated to the free symplectic action of T 2 on KT: T(KT) KT

τ

/ θ = ET 2 × 2 T(KT) T

i

/ ET 2 × 2 KT T

/ BT 2

By our standing assumption, the action preserves a compatible almost complex structure, so the bundle map gives i∗ (c1 (θ )) = c1 (KT). Now, because the action is free, we have the following commutative diagram with q a

8.4

The Chas-Sullivan loop product on loop space homology

homotopy equivalence: i

/ ET 2 × 2 KT KT K T KK KK KK KK q p K% 2 T = KT/T 2 p

Now, the bundle T 2 → KT → T 2 is classified by the top class in H 2 (T 2 ; Z), so p∗ (H 2 (T 2 ; Z)) = 0. Since q is a homotopy equivalence, we therefore also have i∗ (H 2 (ET 2 ×T 2 KT; Z)) = 0. Because c1 (θ ) ∈ H 2 (ET 2 ×T 2 KT; Z), we have c1 (KT) = i∗ (c1 (θ)) = 0.

8.4 8.4.1

The Chas-Sullivan loop product on loop space homology The classical intersection product

Let N be a compact connected oriented n-dimensional smooth manifold and let : N → N × N be the diagonal embedding. Denote by T a tubular neighborhood of (N) and by ∂T its boundary. The exponential map induces a diffeomorphism between T and the normal disk bundle Dν to (N) that restricts to a diffeomorphism between ∂T and the associated sphere bundle Sν. Let p : Dν → N denote the projection. Since N is oriented, there is an orientation class ON ∈ H n (Dν, Sν) such that the cap ∼ =

product with ON induces an isomorphism θ : H∗ (Dν, Sν) → H∗−n (N): θ (x) = H∗ (p)(ON ∩ x). The homology intersection product on N with coefficients in a field lk is the map Hk (N) ⊗ Hl (N) → Hk+l−n (N) obtained as the following composition of maps ∼ =

Hk (N) ⊗ Hl (N) → Hk+l (N × N) → Hk+l (N × N, F(N, 2)) ∼ exc−1 θ = −→ Hk+l (T, ∂T) −→ Hk+l (Dν, Sν) −→ Hk+l−n (N) ,

where exc denotes the excision isomorphism and F(N, 2) denotes the complement of the diagonal (N). In cohomology, we can take the cup product with ON to obtain an isomorphism θ : H ∗−n (N) → H ∗ (Dν, Sν): θ (x) = H ∗ (p)(x) ∪ ON . With coefficients in a field lk, the intersection coproduct is then given by the

341

342


composition of maps θ

∼ =

H ∗−n (N) −→ H ∗ (Dν, Sν) −→ H ∗ (T, ∂T) ∼ exc−1 ∗ = −→ H (N × N, F(N, 2)) → H ∗ (N × N) → H ∗ (N) ⊗ H ∗ (N).

We observe that, since the intersection coproduct is an isomorphism for ∗ = 2n, mapping fundamental class to fundamental class, it is a shriek map by Corollary 8.6, and therefore is multiplication by the diagonal cohomology class (see Lemma 8.17).

8.4.2

The Chas–Sullivan loop product

In [53], M. Chas and D. Sullivan defined a product on the desuspension of the homology (with coefficients in a field lk) of the free loop space of N:

H∗ (LN; lk) = H∗+n (LN; lk) . The product, called the loop product, is defined at the chain level using both the intersection product on the chains on N and the loop composition. We now give a brief description of the Chas–Sullivan definition without going into details. We present a more homological version afterwards. The loop product makes the homology of the free loop space a graded commutative and associative algebra [53]. More precisely, let p : LN → N denote the map that associates to a loop its base point. Let σ : p → LN and τ : q → LN be chains such that their projections in N, pσ and pτ , are transverse. We then define a space E = {(s, t, n) ∈ p × q × N | pσ (s) = pτ (t) = n} as the pullback of the diagram / N

E

p × q

(pσ ,pτ )

/ N×N

Note that the composition of the loops σ (s) and τ (t) is defined for (s, t, n) ∈ E. We thus get a map E → LN by mapping (s, t) to the composition of the loops σ (s) and τ (t). Since N has codimension n in N × N, when σ and τ are transverse, then E has codimension n in p × q . Chas and Sullivan then used this construction to define their loop product, Hp (LN) ⊗ Hq (LN) → Hp+q−n (LN) .

8.4


Let’s now consider a more homological version of the loop product for a compact connected smooth orientable manifold N. Our presentation looks like the homological presentation given above for the intersection product. For Z ⊂ N × N, we denote by LZ the subspace of L(N × N) = LN × LN consisting of loops with base point in Z. For instance L(N) = LN ×N LN, and the composition of loops defines a map that will be very useful later: µ : LN ×N LN → LN . Recall that (T, ∂T) is diffeomorphic to the pair (Dν, Sν) formed by the normal disk bundle and the normal sphere bundle to the embedding . Let p∗ Dν and p∗ Sν denote the pullbacks over LN ×N LN of the disk and sphere bundles Dν and Sν along the projection p : LN ×N LN → (N). A point in p∗ Dν is a pair (c, v) with c ∈ LN ×N LN and v ∈ (Dν)c(0) . Applying the exponential map to v gives a geodesic u(t) = expc(0) (tv). Let u(t) ¯ denote the inverse path, u(t) ¯ = u(1 − t). The correspondence (c, v) → u¯ · c · u gives a homotopy equivalence p∗ Dν → LT , which induces, by restriction, a homotopy equivalence p∗ Sν → L∂T . ¯ N ∈ H n (p∗ Dν, p∗ Sν) the pullback of the orientation Denote now by O n ¯ N defines an isomorphism class ON ∈ H (Dν, Sν). The cap product with O H∗ (L(N) ) ∼ = H∗+n (p∗ Dν, p∗ Sν) . By excision and homotopy equivalence we also have the following isomorphisms: ∼ =

∼ =

H∗ (L(N × N), LF(N,2) ) ← H∗ (LT , L∂T ) → H∗ (p∗ Dν, p∗ Sν) . Definition 8.33 The (Chas–Sullivan) loop product on H∗ (LN) is the composition H∗ (LN) ⊗ H∗ (LN) → H∗ (LN × LN) → H∗ (LN × LN, LF(N,2) ) µ ∼ = H∗ (p∗ Dν, p∗ Sν) ∼ = H∗−n (L(N) ) −→ H∗−n (LN) .

Definition 8.34 The dual of the loop product is the composition αN ◦ H ∗−n (µ), where ⎧ ⎨ H ∗−n (µ) : H ∗−n (LN) → H ∗−n (L (N) ) , ⎩ αN : H ∗−n (L(N) ) ∼ = H ∗ (LN × LN, LF(N,2) ) → H ∗ (LN × LN) .

343

344


8.4.3

A rational model for the loop product

In this section we will suppose that N is 2-connected. Suppose (∧X, d) is a minimal model for N, and let (A, d) be a Poincaré duality model for N. We then have a quasi-isomorphism ϕ : (∧X, d) → (A, d). A model for the free loop space fibration p : LN → N has been described in Section 5.2 (where, ¯ = sX): for the sake of simplicity, we write X ¯ D) , (∧X, d) → (∧X ⊗ ∧X,

D(x) ¯ = −sd(x) .

The pullback ¯ D) ¯ D) = A ⊗∧X (∧X ⊗ ∧X, (A, d) → (A ⊗ ∧X, is also a model for the projection p. Since LF(N,2) is the pullback of LN×LN along the inclusion i : F(N, 2) → N × N, and since the projection π : A ⊗ A → A ⊗ A/(DA ) is a model for i (see Theorem 8.18), a model for the diagram / LN × LN

LF(N,2) F(N, 2)

i

/ N×N

is given by the square

¯ D) ⊗ (A ⊗ ∧X, ¯ D) /(DA ) o (A ⊗ ∧X, O (A ⊗ A)/(DA ) o

π

¯ D) ⊗ (A ⊗ ∧X, ¯ D) (A ⊗ ∧X, O

π

A⊗A

where π is the map ¯ D)⊗2 → (A, d)⊗2 /(DA ) ⊗A⊗A (A ⊗ ∧X) ¯ ⊗2 . (A, d)⊗2 ⊗A⊗A (A ⊗ ∧X, This shows that a model for the injection APL (L(N × N), LF(N,2) ) → APL (L(N × N)) is given by the injection of Ker π by the map µDA , which is the multiplication by DA (see Subsection 8.1.4): ¯ D)⊗(A ⊗ ∧X, ¯ D) µDA ⊗ 1 : (s−n A) ⊗A⊗A (A ⊗ ∧X, ¯ D)⊗2 . → (A ⊗ A)A⊗A (A ⊗ ∧X, From the definition of αN , we deduce the following result.

8.4


Proposition 8.35 ([94]) The map αN is the map induced in cohomology by µDA ⊗ 1. As we saw in Section 5.9, another convenient model for the free loop ¯ D). This space is given by the Hochschild complex B(A) = (A ⊗ T(sA), entails the following version of Proposition 8.35. Proposition 8.36 ([94]) The map αN is the map induced in cohomology by the following multiplication by DA : ¯ ¯ (A ⊗ T(sA)) µDA ⊗ 1 : (s−n A) ⊗A⊗A (A ⊗ T(sA))⊗ ¯ ⊗ (A ⊗ T(sA)) ¯ . → (A ⊗ T(sA)) We now give a model for the multiplication µ : LN ×N LN → LN. Recall ¯ is a differential coalgebra whose comultiplication is defined by that T(sA) ∇([a1 | · · · |aq ]) =

q [a1 | · · · |ai ] ⊗ [ai+1 | · · · |aq ] . i=0

Then we see that ¯ ⊗ T(sA), ¯ D) = A ⊗A⊗A B(A) ⊗ B(A) 1 ⊗ ∇ : B(A) → (A ⊗ T(sA) is a morphism of differential A-modules. Proposition 8.37 The morphism H ∗ (µ; Q) : H ∗ (LN; Q) → H ∗ (L(N) ; Q) is the map induced in cohomology by the morphism 1 ⊗ ∇. Proof The composition of loops µ : L(N) → LN is the pullback morphism in the diagram µ

L(N) _ _ _ _ _ _ _ _ _ _ _ _ _/ LN GG MMM GG MMM GG MMM GG G# M& ϕ¯ / N [0,1] N [0,1] × N [0,1] s N NN NNN NNN NNN NN' N×N×N

ϕ

r N H HH HH HH HH # / N×N

345

346


with ϕ(a, b, c) = (a, c), s(c, c ) = (c(0), c(1), c (1)) and r(c) = (c(0), c(1)). The model for this diagram is 1⊗∇

AO o_gPPP_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ B(A) O eKK PPP KK PPP KK PPP KK PP K θ o B(A, A, A) ⊗ B(A, A, A) B(A, A, A) O

O

A hPPP PPP PPP PPP PP m A⊗A⊗A o

A eLL LLL LLL LLL m θ

A⊗A

where A = A ⊗A⊗A (B(A) ⊗ B(A)), B(A, A, A) denotes the double bar construction (see Section 5.9), the map m denotes the standard multiplication,

θ (a[a1 | · · · |an ]a ) =

n

a[a1 | · · · |ai ] ]1 ⊗ 1[ [ai+1 | · · · |an ]a ,

i=0

= a ⊗ 1 ⊗ a . The model for the composition of loops and θ¯ (a ⊗ is thus obtained by taking the pushout map in this diagram: that is, the map 1 ⊗ ∇. a )

8.4.4

Hochschild cohomology and Cohen–Jones theorem

Recall that the Hochschild cohomology of a graded differential algebra (A, d), HH ∗ (A, A), is the homology of the complex HomA (B(A), A). This homology is a graded algebra whose product " is defined for f , g ∈ ¯ A) = HomA (B(A), A) by Hom(T(sA), f ⊗g ∇ m ¯ −→ ¯ ⊗ T(sA) ¯ −→ f " g : T(sA) T(sA) A ⊗ A −→ A .

Using the isomorphism ψA : A → s−n A∨ , we obtain, by duality, a coproduct on the Hochschild homology of A, HH∗ (A). The coproduct has degree n and is the map induced in cohomology by the composition ψA ⊗1

B(A) = A ⊗A B(A) −→ s−n A∨ ⊗A B(A)

s−n " −n ¯ A) −→ ¯ A) ⊗ Hom(T(sA), ¯ A) s = s−n Hom(T(sA), Hom(T(sA), ψ ⊗2

A ¯ ⊗2 ←− ¯ ⊗2 = sn B(A)⊗2 . sn (A ⊗ T(sA)) = s−n (A∨ ⊗ T(sA))

8.4


We are now ready to prove the theorem of Cohen and Jones. Theorem 8.38 ([62], [191], [94]) The isomorphism H ∗ (LN; Q) ∼ = HH∗ (A) is an isomorphism of coalgebras. By duality, the loop algebra H∗ (LN; Q) is isomorphic as an algebra to the Hochschild cohomology algebra HH ∗ (A, A). Proof The theorem is a direct consequence of the commutativity of the following diagrams (the first one coming from Proposition 8.37 and the second one being a combination of Proposition 8.36 and Lemma 8.17). µ

H ∗ (LN; Q) O

/ H ∗ (L(N) ; Q) O ∼ =

∼ =

H ∗ (B(A))

H ∗ (1⊗∇)

∼ =

H ∗ (ψA ⊗1)

H ∗ (s−n A∨ ⊗A B(A))

s−n H ∗ (L(N) ; Q) O

∼ =

H ∗ (1⊗∇)

/

H ∗ (ψA ⊗1)

H ∗ (s−n A∨ ⊗A⊗2 B(A)⊗2 )

αN

/ H ∗ (L(N × N; Q)) O ∼ =

∼ =

H ∗ (s−n A ⊗A⊗2 B(A)⊗2 ) H ∗ (s−n ψA ⊗1)

/ H ∗ (A ⊗ ⊗2 B(A)⊗2 ) A

H ∗ (µDA ⊗1)

∼ =

∼ =

H ∗ (s−2n A∨ ⊗A⊗2 B(A)⊗2 )

/ H ∗ (A⊗2 ⊗ ⊗2 B(A)⊗2 ) A

H ∗ (m∨ ⊗1)

/

H ∗ (ψA ⊗1)⊗2

H ∗ ((s−n A∨ )⊗2 ⊗A⊗2 B(A)⊗2 )

Since a quasi-isomorphism of differential graded algebras, ϕ : (A, d) → (B, d), induces an isomorphism of Hochschild algebras HH ∗ (A, A) ∼ = HH ∗ (B, B), we can choose A = C ∗ (N; Q), A = APL (N) or A = (∧V, d), where (∧V, d) is the minimal model of N, in Theorem 8.38. We thus have the following isomorphisms of algebras: H∗ (LN; R) ∼ = HH ∗ (ADR (N), ADR (N)) , H∗ (LN; Q) ∼ = HH ∗ ((∧V, d), (∧V, d)) .

347

348


8.4.5

The Chas-Sullivan loop product and closed geodesics

Let (LN)a denote the subspace of LN consisting of curves of length less than or equal to a. Following [110], we define the critical value cr(α) of a homology class α ∈ Hq (LN; Z) to be the number

cr(α) = inf{ a | α ∈ Im Hq ((LN)a ; Z) → Hq (LN; Z) } . Denoting the loop product by •, Goresky and Hingston prove the Theorem 8.39 ([110]) The critical value behaves well with respect to the loop product: cr(α • β) ≤ cr(α) + cr(β). Moreover, we have the following nilpotency result. Theorem 8.40 ([110]) If all closed geodesics are nondegenerate, then every homology class α ∈ H∗ (LN; Z) is level-nilpotent; that is, there is an integer r depending on α such that cr(•r α) < r cr(α). This level-nilpotency of the homology classes for the loop product begs the question of whether the classes are nilpotent in the usual sense. In order to investigate this question, we consider the transverse intersection of a homology class in LN with N viewed as a submanifold of LN. This transverse intersection induces a morphism of algebras I : H∗ (LN; Q) → H∗ (N; Q) [53] that can be computed in the following way. Denote by (∧V, d) the minimal model for N, by (∧V ⊗ ∧sV, δ) the minimal model for LN, and by ϕ : (∧V, d) → (A, d) a finite dimensional model for M with A>n = 0 and An = Qω. Then we have a quasi-isomorphism ϕ ⊗ id : (∧V ⊗ ∧sV, δ) → (A ⊗ ∧sV, δ) = (A, d) ⊗(∧V,d) (∧V ⊗ ∧sV, δ) . Now, the multiplication by ω, α → α · ω, induces a map I ∗ : H ∗ (N; Q) = ∧sV → H ∗ (A ⊗ ∧sV, δ) = H ∗ (LN; Q). Lemma 8.41 ([93]) The map I ∗ is dual to the intersection map I. This computation gives a nilpotence result analogous to the result above of Goresky and Hingston, Theorem 8.40. Theorem 8.42 ([93]) The image of the intersection map I : H∗ (LN; Q) → H∗ (N; Q) is a finitely generated algebra, and its kernel is a nilpotent algebra.

8.4


Exercises for Chapter 8 Exercise 8.1 Let G be a connected Lie group. Show that the loop algebra H∗ (LG; Q) is isomorphic to the tensor product H∗ (G; Q) ⊗ H∗ (G; Q), where H∗ (G; Q) is equipped with the intersection product, and H∗ (G; Q) has the Pontryagin product. Exercise 8.2 Use the Lambrechts–Stanley model Bl(A, B) (see Definition 8.27) to obtain the rational homotopy type of the blow-up of a point in a manifold M. Exercise 8.3 A result of F. Peterson says that, in particular, if a real n-dimensional manifold has torsionfree integral cohomology, then any complex 2n-bundle is complex trivial if and only if its Chern classes vanish. (Note that we are not saying the bundle is complex parallelizable, for this is a restrictive concept requiring analyticity of sections.) Show that we can see the triviality of the tangent bundle of KT directly from the following. p

1. Consider the principal bundle T 2 → KT → T 2 . Show that T(T 2 ) is trivial as a complex bundle because c1 (T 2 ) = χ(T 2 ) = 0 and the trivialization is accomplished by left translation on T 2 . Therefore, the induced action of T 2 on T(T 2 ) = T 2 × C is given by Ly (x, s) = (yx, s). 2. Show that the bundle along the fibers is given by θ = KT ×T 2 (T 2 × C) = (KT ×T 2 T 2 ) × C = KT × C. Hint: to see that KT ×T 2 T 2 = KT is an equivariant homeomorphism, define the inverse maps: λ : KT ×T 2 T 2 → KT, λ([x, g]) = xg; τ : KT → KT ×T 2 T 2 , τ (x) = [x, e]. Hence, θ is a trivial complex bundle. 3. Now show that T(KT) ∼ = θ ⊕ p∗ (T(T 2 )) and therefore it is trivial. Because the tangent bundle is complex trivial, all Chern classes vanish.

349

9

A Florilège of geometric applications

This chapter is a survey on the types of models which arise when studying configuration spaces, smooth algebraic varieties, function (or mapping) spaces, and arrangements. We also give a brief introduction to two subjects, Gelfand–Fuchs cohomology and iterated integrals, which provided unexpected connections between models and geometric analysis at the very dawn of rational homotopy theory. Here we explain the relevant models and then refer to the appropriate literature for details or proofs. Section 9.1 is a quick survey on the rational homotopy type of configuration spaces. When M is a manifold, the configuration space of k points in M is the space F(M, k) = {(x1 , . . . , xk ) ∈ Mk | xi = xj for i = j }. We recall, in particular, how to compute the Betti numbers and the ranks of the homotopy groups of F(M, k). The main problem centers around knowing if the rational homotopy type of F(M, k) depends only on the rational homotopy type of M. By the work of Kriz and Totaro, this is true for complex projective varieties and, by the work of Lambrechts and Stanley, this is true when k = 2 and M is a 2-connected compact manifold. Further, we discuss certain chain complexes giving the rational cohomology of unordered configuration spaces. Next, in Section 9.2, we consider arrangements of hyperplanes and affine subspaces and their accompanying models. The main problem here is to understand the topology of the complement of the arrangement. This is directly related to configuration spaces because the configuration space of k points in R2 is the complement of the arrangement formed by the complex hyperplanes zi = zj . Here we describe the atomic model of Yuzvinsky and give concrete examples. We have seen in Theorem 4.43 that compact Kähler manifolds are formal spaces. We may then wonder if the minimal model of a smooth algebraic manifold also has some special properties. In Section 9.4, with this objective in mind, we first use the existence of a pure Hodge structure on the cohomology of a compact Kähler manifold to give another proof of formality.

9.1

Configuration spaces

In [69], Deligne extended the notion of pure Hodge structure to that of mixed Hodge structure and proved that the cohomology of any complex algebraic manifold carries such a structure. Here, we discuss briefly the result of Morgan (see [201]) which says that the minimal model of a complex algebraic manifold also possesses such a structure. Some homotopy consequences are then given. Spaces of mappings, and, more generally, spaces of sections of fibrations are very useful in geometry. The free loop space is an example of a space of mappings, and the space of sections of the free loop space fibration p : LM → M is the loop space on the topological monoid aut(M). By work of D. McDuff [187], the space of finite parts of M with labels in a space X can be identified with the space of sections of a sphere bundle over M. We explain these ideas in Subsection 9.5.1 and show how to compute the rational homotopy type of the space of sections of a given fibration. Finally, the Gelfand–Fuchs cohomology can be interpreted as the cohomology of the space of sections of a certain bundle. We then obtain explicit computations of the Gelfand–Fuchs cohomology in Subsection 9.5.4. K. Chen’s theory of iterated integrals is another geometric way to obtain rational homotopy information for manifolds. Since this theory has been used in many circumstances in geometry, we have included a presentation of the theory in Section 9.6. This gives us the opportunity to make precise the links with the Sullivan approach. Section 9.7 contains a list of cohomological conjectures that have appeared in different parts of the book and that we present here together with their interrelations.

9.1


Let M denote an m-dimensional manifold. The space of ordered configurations of k points in M is the space F(M, k) = { (x1 , . . . , xk ) ∈ Mk | xi = xj for i = j } . When two manifolds are homeomorphic, then their configuration spaces are also homeomorphic. The natural question is then: When M and N are homotopy equivalent, is it true that F(M, k) and F(N, k) are also homotopy equivalent? By Example 9.1 below, F(Rn , 2) Sn−1 , so we see that the answer is no in general. Now, if M is compact and nonsimply connected, then the answer to this question is also no as shown by R. Longoni and P. Salvatore [176]. On the other hand, Levitt has proven that the answer is yes for compact 2-connected manifolds when k = 2 [172], and M. Aouina and J. Klein have shown that the homotopy type of some suspension of F(M, k) depends only

351

352

9 : A Florilège of geometric applications

on the homotopy type of M [13]. For general compact simply connected manifolds, the problem remains open. In this section, we will be concerned with the rational homotopy of configuration spaces. Of course, here we want to know if the rational homotopy type of F(M, k) depends only on the rational homotopy type of M. In Subsection 8.1.5, we show by a theorem of Lambrechts and Stanley, that this is true for k = 2 when the manifold is 2-connected. We will also show how to get the rational homotopy groups and the rational Betti numbers of F(M, k) directly from a model of M. In case M is a complex projective manifold, a model for F(M, k) can be derived from a model for M. Now, in certain cases we can obtain homotopy equivalences of F(M, k) with well-known spaces. For instance, we have the following. Example 9.1 When M admits a multiplication, then the map (x, y) → (x, x−1 y) induces a homeomorphism between F(M, 2) and M × M\{e}. For instance F(Rn , 2) ∼ = Rn ×(Rn \{0}) Sn−1 . In the same way, F(S1 ×S1 , 2) S1 × S1 × (S1 ∨ S1 ). Example 9.2 When M is a sphere Sn , n ≥ 2, then F(M, 2) Sn and F(M, 3) has the homotopy type of the tangent sphere bundle to Sn . The map from the sphere bundle to F(M, 3) maps (x, v) to (x, −x, expx (v)).

9.1.1

The Fadell–Neuwirth fibrations

Let M be a manifold, and let q1 , . . . , qn be n distinct fixed points in M. For i ≤ n, let Qi = {q1 , . . . , qi }. Observe that the space M\Qi is an open manifold for i ≥ 1, and we have a map π : F(M\Qn−1 , k) → M\Qn−1 sending a k-tuple onto its first component: π(x1 , . . . , xk ) = x1 . Fadell and Neuwirth prove the following result that is basic in the theory. Theorem 9.3 ([83]) The projection π is a locally trivial fiber bundle with fiber F(M\Qn , k − 1). Moreover, when n ≥ 2, the fiber bundle admits a section. Clearly, the fiber over the basepoint qn is F(M\Qn , k−1). Also, the section is easy to construct. In M, we first choose an open n-dimensional disk D around q1 , of radius 1 for some metric, and we suppose that D does not contain any of the points q2 , . . . , qn−1 . We now choose k − 1 distinct points y1 , . . . , yk−1 in D on a sphere of radius 1/2 around q1 . We then define a section σ of π by the following process. When x ∈ D, we put σ (x) = (x, y1 , . . . , yk−1 ). When x ∈ D, then we put σ (x) = (x, |x|y1 , . . . , |x|yk−1 ), where we denote the distance between q1 and x by |x|.

9.1

9.1.2


The rational homotopy of configuration spaces

The homotopy type of the manifold M\Qk is easy to describe. First, since the group of homeomorphisms of M acts k-transitively (i.e. any two given sets of k distinct points can be mapped one onto another by a single homeomorphism), we can suppose that the k points belong to an open set homeomorphic to an m-dimensional open disk D. The space D minus the points has the homotopy type of a wedge of k spheres Sm−1 , one of them being the boundary of the disk. Therefore, if k ≥ 2, the space M\Qk has the homotopy type of the wedge of M\Q1 with a wedge of k − 1 copies of the sphere Sm−1 ; M\Qk M\Q1 ∨ (∨k−1 Sm−1 ). Proposition 9.4 ([90]) If the cohomology algebra H ∗ (M; Q) requires at least two generators, then we have an isomorphism π∗ F(M, k) ⊗ Q ∼ = ⊕k−1 i=0 π∗ (M\Qi ) ⊗ Q . Proof We have the following sequence of fibrations F(M\Q1 , k − 1)

F(M\Q2 , k − 2)

···

M\Qk−1

F(M, k)

F(M\Q1 , k − 1)

···

F(M\Qk−2 , 2)

M

M\Q1

···

M\Qk−2

All the fibrations except maybe the first one admit a section and so their long exact homotopy sequences split into short exact sequences. Recall from Theorem 3.3 that the Lie algebra π∗ ((M\Q1 )) ⊗ Q admits a filtration such that the graded associated Lie algebra is the free product of π∗ (M) ⊗ Q with a free Lie algebra on one generator. Hence, the center of π∗ ((M\Q1 )) ⊗ Q is zero. The same is also true for the center of π∗ ((M\Qi )) ⊗ Q for i ≥ 2 since, by the above description of M\Qi , we have π∗ ((M\Qi )) ⊗ Q ∼ = (π∗ ((M\Q1 )) ⊗ Q) L(x2 , . . . , xi−1 ). Therefore, by induction on the sequence of fibrations above, we find that the center of π∗ (F(M\Q1 , k − 1)) ⊗ Q is zero. Now, we know that the image of the connecting map in the exact rational homotopy sequence for a fibration F → E → B is contained in the center of π∗ (F) ⊗ Q, so the long exact homotopy sequence of the first fibration also splits into short exact sequences. This yields the result.

353

354


9.1.3

The configuration spaces F(Rn , k)

The Betti numbers of F(Rn , k) can be derived from the Fadell–Neuwirth fibrations. Indeed, we have the Proposition 9.5 The integral homology groups of F(Rn \Qm , k) are free abelian groups and we have isomorphisms of groups n−1 H∗ F(Rn \Qm , k) ∼ ). = ⊗k−1 j=0 H∗ (∨m+j S

In particular, n−1 ). H∗ F(Rn , k) ∼ = H∗ F(Rn \Q1 , k − 1) ∼ = ⊗k−1 j=1 H∗ (∨j S

Proof Suppose n > 2. We prove the result by induction on k. For k = 1, we have H∗ F(Rn \Qm , 1) ∼ = H∗ (∨m Sn−1 ) . So suppose the result is true for k−1 points, and any m. Then, for degree reasons, the Serre spectral sequence of the Fadell–Neuwirth fibration collapses at the E2 -level and we have isomorphisms H∗ F(Rn \Qm , k) ∼ = H∗ F(Rn \Qm+1 , k − 1) ⊗ H∗ (Rn \Qm ) n−1 ∼ ) ⊗ H∗ (∨m Sn−1 ) = ⊗k−2 j=0 H∗ (∨m+1+j S n−1 ∼ ) ⊗ H∗ (∨m Sn−1 ) . = ⊗k−1 j=1 H∗ (∨m+j S

The case n = 2 requires more care and is done in detail in [61].

The multiplicative structure of the cohomology of F(Rn , k) has been described by F. Cohen in [61]. Theorem 9.6 Let aij be a sequence of variables of degree n − 1, for i, j = 1, . . . , k. Then we have an isomorphism of algebras, H ∗ (F(Rn , k); Z) ∼ = ∧(aij )/I , where I is the ideal generated by the relations aij − (−1)n aji , a2ij = 0 and the Arnold relations aij ajr + ajr ari + ari aij . This gives the complete rational homotopy type of the configuration spaces of points in Rn because, by a result of Kontsevich, the spaces F(Rn , k) are formal. Theorem 9.7 ([156, Theorem 2]) The spaces F(Rn , k) are formal.

9.1


The case n = 2 was proved previously by Arnold [42]. Indeed, F(R2 , k) is the complement in Ck of the union H of the hyperplanes Hij = { (z1 , . . . , zn ) | zi = zj }. The projection ϕij = zi − zj : Ck \Hij → C\{0} is a homotopy equivalence and this implies that H 1 (Ck \H; R) is generated by the classes of the 1-forms dϕij dz eij = ϕij∗ = . 2πiz 2π iϕij A simple computation then shows that the Arnold relations (A) are satisfied:

A : eij ejr + ejr eri + eri eij = 0 . This defines, by restriction, a morphism of differential graded algebras (∧(eij )/(A), 0) → ADR (F(C, k)) , which is a quasi-isomorphism. This then proves the formality of F(R2 , k).

9.1.4

The configuration spaces of a projective manifold

Independently, and by using different methods, I. Kriz [160] and B. Totaro [254] gave a model for the configuration spaces F(M, k) when M is a complex projective manifold. In order to describe this model, first let pi : Mk → M denote the projection on the i-th component and let pij : Mk → M2 denote the projection on components i and j. Since M is a manifold, we j can find a graded basis {ai , a j } of H ∗ (M; Q) such that ai ∪ a j = δi ω where j

ω denotes a fundamental class and δi is the Kronecker delta. The class DM = i (−1)|ai | ai ⊗ a i ∈ H ∗ (M × M; Q) is called the diagonal class and we set Dij = p∗ij (DM ) ∈ H ∗ (M; Q)⊗k . Then we have Theorem 9.8 Let Mm be a complex projective manifold. Then a model for F(M, k) is given by the differential graded algebra (H ∗ (M; Q)⊗k ⊗ ∧(xij , i, j = 1, . . . k) / I, d) , where |xij | = m−1, d(xij ) = Dij and the ideal I is generated by the relations ⎧ ⎪ ⎪ x = xji , ⎪ ⎨ ij xij xjr + xjr xri + xri xij , ⎪ ⎪ ⎪ ⎩ p∗ (x) · x = p∗ (x) · x . i

ij

j

ij

355

356


Note that, in this description, we recover the Arnold relations. Also note that the compatibility of the differential with the Arnold relations comes from Poincaré duality. Once again, we have a situation where we obtain a complete description of the rational homotopy type. Now suppose that Mm is any compact, simply connected manifold and let (A, d) denote a Poincaré duality model for M with diagonal class DA (see Definition 8.16). Then we can form a new commutative differential graded algebra by replacing (H ∗ (M; Q), 0) by (A, d) in the model of Kriz and Totaro. We let pi : A → A⊗k denote the injection pi (a) = 1⊗· · ·⊗a⊗· · ·⊗1 with the element a in the ith position. In a similar way pij : A⊗2 → A⊗k denotes the map that sends a ⊗ b to the k-tuple x1 ⊗ · · · ⊗ xk with xi = a, xj = b and the other xi equal to 1. We can then form the cochain algebra G(M, k) = (A⊗k ⊗ ∧(xij , i, j = 1, . . . k) / I , D) , where D(a) = d(a), |xij | = m − 1, D(xij ) = pij (DA ) and the ideal I is generated by the relations ⎧ 2 m ⎪ ⎪ ⎨ xij = (−1) xji , xij = 0, xij xjr + xjr xri + xri xij , ⎪ ⎪ ⎩ p (x)x = p (x)x . i

ij

j

ij

Conjecture 9.9 G(M, k) is a model for the configuration space F(M, k). The conjecture is true for complex projective manifolds. By a result of Lambrechts and Stanley, it is also true for 2-connected manifolds when k = 2 (Subsection 8.1.5). Some other results go in the direction of the conjecture; for instance the cohomology of G(M, k) is the right cohomology. Theorem 9.10 There is an isomorphism of graded vector spaces H ∗ (G(M, k)) ∼ = H ∗ (F(M, k); Q) . This result is essentially the work of Bendersky and Gitler [24]. The present presentation comes from a re-writing due independently to B. Berceanu, M. Markl and S. Papadima [26] on one hand, and to Y. Félix and J.-C. Thomas [90] on the other. Note that the permutation group k acts freely on F(M, k). The quotient Ck (M) = F(M, k)/ k is called the space of unordered configurations of k points in M. This construction gives a lot of interesting spaces. For instance, when M is the space R2 , then π1 (Ck (M)) is the braid group Bk on k generators. Of course, even if M is simply connected, Ck (M) is not simply connected, and so its minimal model gives less information than we

9.1


might hope. We can, however, compute the rational cohomology of Ck (M) in case the dimension of M is odd. Theorem 9.11 ([31]) If dim M = m is odd, then H ∗ (Ck (M); Q) = ∧k H ∗ (M; Q) . For instance, when M = S3 , a basis of the vector space of cohomology is given by e0 in degree 0 and e3 in degree 3. The only words of length k in ∧(e0 , e3 ) are e0k and e0k−1 e3 . Therefore for any k ≥ 1, H ∗ (Ck (S3 ); Q) = ∧(z) with z in degree 3. In particular, Ck (S3 ) Q S3 . Now let M be a compact simply connected m-dimensional manifold, with m even. Let V = H ∗ (M; Q), W = s−m+1 H ∗ (M; Q) and let DM ∈ V ⊗ V be the diagonal class. We consider the differential graded algebra (∧(V ⊕ W), d) , where d(V) = 0 and d(s−m+1 a) = DM · (1 ⊗ a) ∈ V ⊗ V for s−m+1 a ∈ W. In fact we need to write this last expression as an element of ∧2 V. For that, if DM = i (−1)|ai | ai ⊗ a i , then we have 2 3 1 −m+1 |ai | d(s a) = (−1) ai ∧ (ai · a) , 2 i

where the · means the cohomology multiplication. Example 9.13 gives a concrete example of computation. The algebra ∧(V ⊕ W) can also be equipped with a second gradation (∧(V ⊕ W))k = ⊕2p+q=k ∧p W ⊗ ∧q V . the cohomology decomThen d(∧(V ⊕ W))k ⊂ (∧(V ⊕ W))k . Therefore, poses as a direct sum H ∗ (∧(V ⊕ W), d) ∼ = k Hk∗ (∧(V ⊕ W), d). We then have Theorem 9.12 ([91]) With the above notation, when m is even, there is an isomorphism of graded vector spaces Hk∗ (∧(V ⊕ W), d) ∼ = H ∗ (Ck (M); Q) . Example 9.13 To compute H ∗ (Ck (CP3 ); Q), we have to form the cdga (∧(e0 , e2 , e4 , e6 , x5 , x7 , x9 , x11 ), d)

357

358


where the subscripts indicate the degrees, and d(ei ) = 0, d(x5 ) = e0 e6 +e2 e4 , d(x7 ) = e2 e6 + (1/2)e42 , d(x9 ) = e4 e6 , d(x11 ) = (1/2)e62 . We define a second (lower) gradation by saying the xi are in degree 2 and the ej in degree 1. The differential preserves the gradation, and Hk∗ (∧(ei , xj ), d) = H ∗ (Ck (CP3 ); Q). The multiplicative structure of H ∗ (Ck (M); Q) has been explained in [88]. As an illustration, we extract from [88] the rational cohomology algebra of Ck (CP2 ): ⎧ ⎪ ⎪ |x| = 2 ; k = 1, 2 ∧x/x3 ⎪ ⎪ ⎨ k=3 ∧(x, y)/(x3 , yx2 ) |x| = 2 , |y| = 3 ; ⎪ ⎪ ⎪ ⎪ ⎩k ≥ 4 |x| = 2 , |y| = 3 . ∧(x, y)/x3

9.2

Arrangements

An arrangement A of linear subspaces in a complex vector space Cn is simply a finite set of linear subspaces of Cn . The associated intersection lattice L(A) is the poset of intersections among subspaces in A ordered with reversed inclusions: x ≤ y if and only if y ⊂ x. For x, y ∈ L(A), the meet x ∧ y is defined to be x ∧ y = ∩{z ∈ L(A) |x ∪ y ⊂ z }, and the join is defined to be the intersection x ∨ y = x ∩ y. With these operations, L(A) is a lattice. If σ is a collection of elements in A, we denote by ∨σ , the join of all the elements in σ . For x ∈ L(A), the rank of x, rk(x), is the maximal length r of a sequence of the form Cn < x1 < x2 < · · · < xr = x, with xi ∈ L(A). When x < y and there is no z such that x < z < y, we write x ≺ y. The lattice L(A) is said to be geometric if each time x ≺ y, then, for each z, either x ∨ z ≺ y ∨ z or x ∨ z = y ∨ z. When a lattice is geometric, the following properties hold: 1. If x < y, all the maximal sequences x < x1 < · · · < xn = y have the same length. 2. For every x, y ∈ L(A), we have rk(x) + rk(y) ≥ rk(x ∨ y) + rk(x ∧ y). The arrangement A is said to be geometric if the lattice L(A) is geometric. The complement, M(A), of an arrangement A is the complement of the union of the elements of A; that is, M(A) = Cn − ∪x∈A x. When A is an arrangement of codimension one subspaces, then the cohomology of M(A) is well known and the space is formal by a result of Brieskorn [42]. When all the subspaces x have codimension at least 2, then M(A) is simply connected and rational homotopy theory applies.

9.2

Arrangements

Example 9.14 Denote by A the arrangement of C2n formed by the subspaces Hi = {(z1 , . . . 5 , z2n ) | z2i−1 = z2i = 0}, for i = 1, . . . , n. Then the projection M(A) → ni=1 C2 \{0} is a homeomorphism and we obtain a homotopy equivalence M(A) (S3 )n . Example 9.15 Let A be an arrangement in Cn , and consider the injection j of Cn into Cn+1 as the hyperplane zn+1 = 0. The image of A by the map j is an arrangement A in Cn+1 and we clearly have M(A ) 2 M(A). Example 9.16 Consider in C6 the arrangement A = {H1 , H2 , H3 , L} where Hi = {(z1 , . . . , z6 ) | z2i−1 = z2i = 0}, and L = {(z1 , . . . , z6 ) | z1 = · · · = z6 }. This arrangement is not geometric. The chain C6 < L is maximal, but taking intersections with H1 produces the non-maximal chain H1 < H1 ∨ L = 0 (where, for instance, the subspace K = {(z1 , . . . , z6 ) | z1 = z2 = z3 = 0} ∈ L(A) lies between 0 and H1 ). Let A be an arrangement of linear subspaces in Cn . Since we are interested in the complement of the union of the elements of A, we suppose that A does not contain any two elements x, y such that x ⊂ y. This implies that the rank of any element x ∈ A is 1. The relative atomic complex of A is the complex (D, d) defined by Yuzvinsky as follows [267]. The complex D is the Q-vector space generated by all the subsets σ = (x1 , x2 , . . . , xk ) of A. We write |σ | to denote the number of elements in σ , and we choose an ordering on the elements of A, so that each sequence is always written following this ordering. The differential of σ = (x1 , x2 , . . . , xk ) is defined by (−1)j σ \{xj } , d(σ ) = j∈J(σ )

where J(σ ) = { j ∈ {1, . . . , k} | ∨ σ = ∨(σ \{xj }) }. The degree of σ is defined by deg σ = 2 codim (∨σ ) − |σ | . Finally we can define a multiplication on D as follows. For two sequences σ , τ , take (−1)ε(σ ,τ ) σ ∪ τ if codim ∨σ + codim ∨τ = codim ∨(σ ∪ τ ) σ ·τ = 0 otherwise. Here “codim” denotes the complex codimension and ε(σ , τ ) the sign of the permutation that must be applied to σ ∪ τ to put the elements in the chosen linear order.

359

360


With those choices, (D, d) is a cochain complex [267]. The following result, due to Yuzvinsky, gives the link between the geometry and the topology of an arrangement [267]. Theorem 9.17 The relative atomic model of Yuzvinsky is a model of the complement M(A). Example 9.18 Let’s describe the relative atomic model associated to the arrangement of Example 9.16. By definition we have d(L) = d(H1 ) = d(H2 ) = d(H3 ) = d(H1 , H2 ) = d(H1 , H3 ) = d(H2 , H3 ) = 0 , d(H1 , H2 , H3 ) = d(L, H1 ) = d(L, H2 ) = d(L, H3 ) = 0 d(L, H1 , H2 ) = (L, H2 ) − (L, H1 ), d(L, H1 , H3 ) = (L, H3 ) − (L, H1 ), d(L, H2 , H3 ) = (L, H3 ) − (L, H2 ), d(L, H1 , H2 , H3 ) = −(H1 , H2 , H3 ) + (L, H2 , H3 ) − (L, H1 , H3 ) + (L, H1 , H2 ). The elements (Hi ) have degree 3 and generate, in cohomology, an exterior algebra ∧(H1 , H2 , H3 ). There are two other cohomology classes: [(L)] in degree 9, and [(L, H1 )] in degree 10. We then consider the cdga (∧(x1 , x2 , x3 ) ⊕ Qy ⊕ Qz, 0), where xi y = xi z = yz = z2 = 0, |xi | = 3, |y| = 9 and |z| = 10. The morphism ϕ : (∧(x1 , x2 , x3 ) ⊕ Qy ⊕ Qz, 0) → (D, d) defined by ϕ(xi ) = (Hi ), ϕ(y) = (L) and ϕ(z) = (L, H1 ), is then a quasi-isomorphism. This shows that M(A) has the rational homotopy type of (S3 ×S3 ×S3 )∨S9 ∨S10 . For geometric arrangements, the following notion is important as we will see in Lemma 9.21. Definition 9.19 A sequence σ = (x1 , x2 , . . . , xk ) of elements of A is said to be independent if, for each k, ∨σ > ∨(σ \{xk )}. Note that subsequences of independent sequences are also independent. Lemma 9.20 If the lattice L(A) is geometric, then a sequence σ = (x1 , . . . , xk ) is independent if and only if rk σ = k.

9.2

Arrangements

Proof Suppose σ is independent. Then each subsequence of σ is independent and we have a sequence of inequalities x1 < x1 ∨ x2 < x1 ∨ x2 ∨ x3 < · · · < ∨σ . Therefore rk σ ≥ k. If rk σ > k, then this sequence is not maximal and there is another element y ∈ A such that for some p (x1 ∨ · · · ∨ xp ) < (x1 ∨ · · · ∨ xp ) ∨ y < (x1 ∨ · · · ∨ xp ) ∨ xp+1 . This will imply that rk (x1 ∨ · · · ∨ xp ∨ xp+1 ) ≥ rk (x1 ∨ · · · ∨ xp ) + 2 > rk (x1 ∨ · · · ∨ xp ) + rk (xp+1 ),

(9.1) (9.2)

in contradiction with the definition of a geometric lattice. Suppose now that rk σ = k and consider the sequence xτ1 ≤ xτ1 ∨ xτ2 ≤ · · · ≤ ∨σ , obtained from σ by some permutation τ of the set { 1, 2, . . . , k }. If all inequalities are strict for any τ , then σ is independent. Otherwise, some inequality is in fact an equality, and, since the rank of σ is k, another part of the sequence extends to the form (xτ1 ∨ · · · ∨ xτr ) < (xτ1 ∨ · · · ∨ xτr ) ∨ y < (xτ1 ∨ · · · ∨ xτr ) ∨ xτr+1 . As above, the existence of such a sequence is impossible for a geometric lattice. It follows directly from the definition of the differential that independent sequences are cocycles. The key point in the geometric case is the following result. Lemma 9.21 ([85]) If the lattice L(A) is geometric, then each cohomology class of (D, d) is represented by a linear combination of independent sequences. Of course, now we want to see what the salient properties of arrangements are with respect to rational homotopy theory. The first property is an analogue of Brieskorn’s result for geometric arrangements [42].

9.2.1

Formality of the complement of a geometric lattice

Theorem 9.22 ([85]) Let A be a subspace arrangement. If L(A) is a geometric lattice, then the space M(A) is formal.

361

362


Proof Define a linear map ψ : D → H ∗ (D, d) by σ →

[σ ] if σ is independent, 0 otherwise.

We first prove that the map ψ is multiplicative. Now, by Lemma 9.20, if σ is not independent, then σ ∪ τ is not independent and ψ(σ ) = ψ(σ · τ ) = 0. We suppose therefore that σ and τ are independent. If σ ∪ τ is not independent, we can suppose there is x ∈ σ such that ∨(σ ∪ τ ) = ∨(σ \{x} ∪ τ ). In that case codim ∨ (σ ∪ τ ) = codim ∨ (σ \{x} ∪ τ ) ≤ codim ∨ (σ \{x}) + codim ∨ τ < codim ∨ σ + codim ∨ τ . Therefore, by definition of the product and of ψ, we have, ψ(σ ∪τ ) = 0 and ψ(σ ) · ψ(τ ) = [σ ] · [τ ] = [σ · τ ] = 0. In case σ , τ and σ ∪ τ are independent, the result follows from the equality ψ(σ ) · ψ(τ ) = [σ ] · [τ ] = [σ · τ ] = ψ(σ · τ ). We now prove that ψ is a morphism of complexes: that is, ψ(dσ ) = 0 for every σ . Write σ = (x1 , . . . , xk ) and σi = σ \{xi }. When all the σi are dependent, ψ(σi ) = 0 for all i, and the result is proved. When all the σi are such that ∨σ = ∨σi are independent, we have 2 3 8 9 i i ψ(dσ ) = ψ (−1) σi = (−1) σi = [dσ ] = 0. i

i

These two cases are in fact the only possibilities. Suppose indeed that ∨σ = ∨σi = ∨σj and all the σi are independent. Then by Lemma 9.20, rk σi = k − 1. Since rk σj = rk σ = rk σi = k − 1, all the σj are also independent. By Lemma 9.21, ψ induces a surjective map in cohomology. The map H(ψ) is clearly injective and therefore ψ is a quasi-isomorphism.

9.2.2

Rational hyperbolicity of the space M(A)

Let A = {x1 , . . . , xq } be a geometric arrangement of subspaces of codimension at least two. When each subspace is transversal to any intersection of the other ones in A, then the complement is a product of odd dimensional spheres. In fact, by a result of G. Debongnie [68], this is the only

9.3

Toric topology

case where M(A) satisfies Poincaré duality. More precisely, Debongnie’s theorem states: Theorem 9.23 ([68]) Let A be a geometric arrangement. Then the following conditions are equivalent: 1. 2. 3. 4.

The rational cohomology of M(A) satisfies Poincaré duality. M(A) is a rationally elliptic space. codim(∩x∈A x) = x∈A codim(x). M(A) is a product of odd dimensional spheres.

Remark 9.24 The assertion (3) ⇒ (4) is easy to understand. Suppose (3) is verified. Let x1 , . . . , xr be the subspaces in A. Then the quotient map p : Cn → Cn /(∩xi ) induces a homotopy equivalence Cn \(∪xi ) → (Cn /(∩xi ))\ ∪ ( xi / ∩ xi )). Hence we can suppose ∩xi = {0}. Write xi = Ker(ϕi : Cn → Cni ). The map ϕ = (ϕ1 , ϕ2 , . . . , ϕr ) : Cn →

r

Cni

i=1

is a homeomorphism which induces a homotopy equivalence r C \ ∪ xi → (Cni \{0}) . n

i=1

5r 2ni −1 ) → Since we also have a homotopy equivalence i=1 (S 5 r n i \{0}), the space M(A) has the homotopy type of a product of ( C i=1 odd dimensional spheres.

9.3

Toric topology

Recently, ideas and tools from rational homotopy theory have proved to be important in the study of toric spaces. Let’s now briefly discuss these notions. Denote C∗ = C − {0}. A toric variety is a normal algebraic variety M containing the algebraic torus (C∗ )n as a Zariski open subset in such a way that the natural action of (C∗ )n on itself extends to an action on M. A fundamental example is given by the so-called moment-angle complex ZK defined for any simplicial complex K in [45]. When K has m vertices, ZK is the complex ∪σ Dσ ⊂ (D2 )m , where Dσ = { (z1 , . . . , zm ) ∈ (D2 )m | |zi | = 1 if i ∈ σ } . Clearly, each ZK is a toric variety. In the sequel, we will only consider moment-angle complexes. The Davis–Januszkiewicz space associated to K,

363

364


DJ(K) is defined by the Borel construction DJ(K) = ZK ×T m ET m . For any coefficient ring R, the cohomology ring H ∗ (DJ(K); R) is isomorphic to the Stanley–Reisner algebra R(K) [45]. Recall that R(K) = R[x1 , . . . , xm ]/IK where |xi | = 2 and the ideal IK is generated by the square free monomials xi1 . . . xir with σ = {i1 , . . . , ir } ∈ K. The space DJ(K) can be shown to be homotopy equivalent to the subspace ∪σ ∈K (BT)σ of (BT)m , with (BT)σ ∼ = (BT)dim σ , see [45]. We then have the following formality result due to Notbohm and Ray. Theorem 9.25 ([212]) The space DJ(K) is rationally formal. Indeed, this formality result is valid for any ring R. As a corollary, a model for DJ(K) is given by (Q(K), 0) and a model for the momentangle complex ZK is given by (Q(K) ⊗ ∧(u1 , . . . , um ), d) with d(ui ) = xi . However, the spaces ZK are not always formal as shown by examples of Denham and Suciu [72]. The spaces ZK are also related to spaces of arrangements. Let K be a simplicial complex on the set 1, . . . , m. We associate to it the complex coordinate arrangement AK = { Lσ | σ ∈ K }, where, for σ = {i1 , . . . , ir }, Lσ = { (z1 , . . . , zm ) ∈ Cm | zi1 = zi2 = · · · = zir = 0 } . Then, we have Theorem 9.26 ([45]) The spaces M(AK ) = Cm \ ∪σ Lσ and ZK have the same homotopy type.

9.4

Complex smooth algebraic varieties

In [201], John Morgan studied models of smooth complex algebraic varieties. A key to his approach is the notion of pure Hodge structure. Definition 9.27 Let V be a Q-vector space. A pure Hodge structure of weight n on V is a finite bigradation on V c = V ⊗Q C, V c = ⊕p+q=n V p,q , such that V p,q = V q,p . Remark 9.28 The existence of a pure Hodge structure of weight n on V is equivalent to the existence of a decreasing filtration F • on V c such that V c = F 0 (V c ) ⊃ F 1 (V c ) ⊃ . . . ⊃ F n (V c ) ⊃ F n+1 (V c ) = {0}

9.4

Complex smooth algebraic varieties

and V c = F p (V c ) ⊕ F n+1−p (V c ), for any p. The correspondence goes as follows. • If F • is given, one sets V p,q = F p ∩ F q . • If the bigradation is given, we set F p (V c ) = ⊕k≥p V k,∗ .

For instance, if M is a connected compact Kähler manifold M, the fact that the different Laplace operators d , ∂ , ∂ coincide implies the existence of a pure Hodge structure of weight k on H k (M; Q), see [71, Section 5]. Theorem 9.29 Let M be a compact connected Kähler manifold. Then the complex minimal model (∧V, d) of M admits a bigradation such that the following properties hold. 1. Let Vk be the subspace of elements of degree k in V. There is a decomposition of Vk into Vk = Ck ⊕ Nk , with d|Ck = 0, d injective on Nk and i,j

i,j

Ck = ⊕i+j=k Ck and Nk = ⊕i+j≥k+1 Nk . 2. The bigradation is extended multiplicatively to (∧V, d) and d(∧V)i,j ⊂ (∧V)i,j . 3. The quasi-isomorphism ρ : (∧V, d) → (Ac (M), d) is compatible with the bigradation Ac (M) = ⊕p,q Ap,q . 4. The bigradation induced on the cohomology of (∧V, d) gives the pure Hodge structures of H(M; C). The proof is modelled along the steps of construction of a minimal model and uses an argument involving the pure Hodge structure on H k (M; C). We refer the reader to [71, page 271] for more details. As a consequence, this result gives a short proof of the formality of M. Corollary 9.30 A compact connected Kähler manifold M is formal. Proof Let ρ : (∧V, d) → (Ac (M), d) be the model of M described above. We define Ci to be the subset of Vi consisting of the elements of degree i and bidegree (r, s) with r + s = i. We denote by Ni a complement of Ci in Vi . The elements of Ni are of degree i and of bidegree (r, s) with r + s > i. Observe that the elements of Ci are cocycles and that the differential d is injective on Ni . Denote by I the ideal of ∧V generated by ⊕i Ni . If z is a cocycle of degree j in I t,u , the class [z] is of degree j and bidegree (t, u) with t + u > j. Since H j (∧V, d) ∼ = H j (M; C) has a pure Hodge structure of weight j, we have [z] = 0 and the ideal I is acyclic. The conclusion is now a direct consequence of Exercise 2.3.

365

366


So, we have a bigradation on the minimal model of a compact Kähler manifold which induces the pure Hodge structures on the cohomology. In order to extend this structure to minimal models of smooth algebraic varieties, we have to replace the notion of pure Hodge structure with that of mixed Hodge structure. The next remark gives the flavor of what a mixed Hodge structure is. Remark 9.31 Consider two smooth projective algebraic varieties X1 and X2 such that X1 ∩ X2 is also a smooth projective algebraic variety. If we want to study X = X1 ∪ X2 , the first elementary tool is the Mayer-Vietoris exact sequence which includes, for instance with k > 1, ...

/ H k−1 (X ∩ X ) 1 2

δ

/ H k (X)

ϕ

/ H k (X ) ⊕ H k (X ) 1 2

/ ...

In H k (X), two types of pure Hodge structures are interfering: 1. a structure of weight k − 1 coming from H k−1 (X1 ∩ X2 ); 2. a structure of weight k coming from H k (X1 ) ⊕ H k (X2 ). To separate them, we introduce a filtration W• on H k (X) defined by: 1. Wk−2 = 0; 2. Wk−1 = Im δ; 3. Wk = H k (X). On the associated graded vector spaces, Grk−1 H k (X) = Im δ and Grk H k (X) = H k (X)/Ker ϕ, we get pure Hodge structures of respective weights k−1 and k. More generally, the definition of mixed Hodge structure goes as follows. Definition 9.32 Let V be a Q-vector space. A mixed Hodge structure on V consists of two filtrations (W• , F • ) such that: 1. W• is an increasing filtration on V, called the weight filtration; 2. F • is a decreasing filtration on V c = V ⊗ C, called the Hodge filtration; 3. the filtration F • induces a pure Hodge structure of weight k on each of the GrW k (V) = Wk /Wk−1 . In [69], [70], Deligne constructed a mixed Hodge structure on the cohomology of any complex algebraic variety. Morgan adapted the notion of mixed Hodge structure to the context of differential algebras and defined the notion of mixed Hodge diagrams, see [201, Definition 3.5]. In particular, any complex smooth algebraic variety gives rise to a mixed Hodge diagram and its model has a bigradation.

9.5

Spaces of sections and Gelfand–Fuchs cohomology

Theorem 9.33 If M is the minimal model of a simply connected, smooth complex algebraic variety, then 1. we can write M = ⊕0≤r,s Mr,s with M0,0 = C, Mr,s Mt,u ⊂ Mr+t,s+u and d Mr,s ⊂ Mr,s ; 2. the minimal model M admits a mixed Hodge structure and the induced mixed Hodge structure on its cohomology coincides with the mixed Hodge structure on the cohomology of the variety. From this, Morgan deduces the following properties. • The homotopy groups πk (X)⊗ Q of a simply connected, smooth complex

algebraic variety are endowed with a mixed Hodge structure coming from the mixed Hodge structure of the variety’s minimal model. (For the nonsimply connected case, we send the reader to [202, Theorem 9.2].) • Not all finite, simply connected CW-complexes are homotopy equivalent to smooth complex algebraic varieties since there exist minimal models of finite, simply connected CW-complexes that cannot have bigradations (see [201, page 196]). Concerning, the existence of mixed Hodge structures on the homotopy groups, a different approach was used by R. Hain. In short, he proved that if A is a connected multiplicative mixed Hodge complex, then the Bar construction on A is a mixed Hodge complex. This structure induces a mixed Hodge structure on the indecomposables that correspond to homotopy groups. The details are in [128].

9.5 9.5.1

Spaces of sections and Gelfand–Fuchs cohomology The Haefliger model for spaces of sections

Let F

/ E j

p

/ B

σ

be a fibration with section σ where B and F are simply connected and B is finite dimensional. Then the space (p) of sections of p that are homotopic to σ is a nilpotent space [203], and a model for (p) has been constructed by Haefliger [126]. Our goal here is to describe this model. The construction starts with a relative minimal model of p ¯ . (∧V, d) → (∧V ⊗ ∧W, D) → (∧W, D)

(9.3)

367

368


We first show that we can always modify this model to obtain a relative minimal model such that a model for the section σ is the map ρ : (∧V ⊗ ∧W, D) → (∧V, d) that is the identity on V and maps W to 0. Since p admits a section, there is a map of cdga’s ρ : (∧V ⊗ ∧W, D) → (∧V, d) that is the identity on V. We now change generators by replacing each w ∈ W by w −ρ(w) and this gives an automorphism of (∧V ⊗∧W, D) that we use to modify the differential D. Clearly, we now have ρ(W) = 0. Also, because we have modified the generators w by elements of ∧V, we ¯ remains unchanged. see that the differential in the fiber, D, We now choose a connected and finite dimensional model (A, d) for (∧V, d),

(∧V, d) −→ (A, d), and we tensor the models in (9.3) with (A, d) over (∧V, d) to obtain a new relative model for the fibration, ¯ . (A, d) → (A ⊗ ∧W, D) → (∧W, D) Let A∨ = Hom(A, Q) be the dual of A. Now denote a graded basis for A by (ai ) and the dual basis of A∨ by (a i ). The duality < , > between A and A∨ means that a i , aj = 1 if i = j and 0 otherwise. We give a i the degree |a i | = −|ai |. Since A is finite dimensional, we can consider the morphism of commutative algebras, θW : A ⊗ ∧W → A ⊗ ∧(A∨ ⊗ W) , defined by θW (a) = a,

θW (w) =

ai ⊗ (a i ⊗ w) .

i

Haefliger proves the existence of a unique differential D on ∧(A∨ ⊗ W) such that θW : (A ⊗ ∧W, D) → (A, d) ⊗ (∧(A∨ ⊗ W), D) is a morphism of cdga’s. Let Q(D) be the linear part of the differential D in (∧(A∨ ⊗ W), D). Denote by I the graded differential ideal of ∧(A∨ ⊗ W) generated by (A∨ ⊗ W)≤0 and Q(D)(A∨ ⊗ W)0 , and form the quotient cochain algebra M = (∧(A∨ ⊗ W)/I, D). The combination of θW and the quotient map (∧(A∨ ⊗ W), D) → M gives a map of cdga’s ϕ : (A ⊗ ∧W, D) → (A, d) ⊗ M . We can now state Haefliger’s theorem [126].

9.5


Theorem 9.34 Let ev : (p) × B → E denote the evaluation map given by (s, b) → s(b). Then the morphism ϕ is a model for ev. In particular M is a Sullivan model for the space of sections (p) homotopic to σ . Notice that the space of functions from X into Y may be identified with the space of sections of the trivial bundle X×Y → X. Therefore, the process above furnishes us with a model for function spaces. We give more details on this construction below. Example 9.35 A model for the space of free maps, Map(Sn , X), can be obtained from the Haefliger model. First of all, a finite dimensional model for Sn is given by the exterior algebra (∧a, 0) if n is odd and ((∧a)/a2 , 0) if n is even. We denote this cdga by E(a). Let (∧W, d) be the minimal model for X. A model for the evaluation map Sn × Map(Sn , X) → Sn × X is then given by ϕ : E(a) ⊗ (∧W, d) → E(a) ⊗ (∧(W ⊕ sn W), D) , where we have written sn W for a ⊗ W. By definition, we have ϕ(x) = x + a · sn x for any x ∈ W. Extend sn to ∧W as a derivation of degree −n. Then we have for any element α ∈ ∧W, ϕ(α) = α + a · sn (α). The compatibility of ϕ with the differentials then gives, for each element x ∈ W, dx + asn (dx) = ϕ(dx) = Dϕ(x) = D(x + asn x) = dx + (−1)n aD(sn x) . Therefore D(sn x) = (−1)n sn (dx). In the case n = 1, we therefore recover the model of the free loop space obtained in Theorem 5.11. Using the Haefliger model, we can obtain information about the rational homotopy groups and rational cohomology of (p) when B is n-dimensional and F is (n + 1)-connected. p

Proposition 9.36 Let F → E → B be a nilpotent fibration and suppose that B is n-dimensional and F is (n + 1)-connected. Then the space (p) is simply connected and we have an isomorphism of graded Lie algebras, πq ((p)) ⊗ Q ∼ = ⊕r≥0 Hom(Hr (B; Q), πq+r (F) ⊗ Q) . The Lie bracket on the right-hand side is obtained by taking the diagonal in homology followed by the bracket in π∗ (F) ⊗ Q: [f , g](x) =

[f (xi ), g(xj )], i,j

if the diagonal is given by (x) =

i,j

xi ⊗ xj .

369

370


In particular, as a graded vector space, we have πq ((p)) ⊗ Q ∼ = ⊕r≥0 H r (B; Q) ⊗ πq+r (F) ⊗ Q . Denote by Map(X, Y, f ) the component of the mapping space Map(X, Y) consisting of the maps homotopic to f : X → Y. The rational homotopy groups of Map(X, Y, f ) have been computed independently by G. Lupton and S. Smith [177], U. Buijs and A. Murillo [46] and J. Block and A. Lazarev [30]. Denote by ϕ : (∧V, d) → (A, d) a model of f with (∧V, d) a Sullivan model for Y. A linear map g : ∧ V → A of degree q is a ϕ-derivation if g(xy) = g(x)ϕ(y) + (−1)q|x| ϕ(x)g(y). Together with the differential D defined by D(g) = dg − (−1)|g| gd, the ϕ-derivations form a complex Der((∧V, d), (A, d), ϕ). Proposition 9.37 Suppose X and Y are simply connected. Then there is an isomorphism of graded vector spaces, ∼ =

H−q Der((∧V, d), (A, d), ϕ) −→ πq Map(X, Y, f ) ⊗ Q. When Y is rationally hyperbolic and X is finite, the space of maps from X into Y has infinite Lusternik–Schnirelmann category and its cohomology is usually very large. For instance M. Vigué-Poirrier proved the following exponential growth law for the Betti numbers of particular mapping spaces. Theorem 9.38 ([260]) Let X M be the space of maps from a compact connected manifold M into a space X that has the rational homotopy type of a wedge of simply connected spheres. Then the Betti numbers of X M exhibit exponential growth; that is, there is an A > 1, such that for k large enough, k i M k i=0 dim H (X ; Q) ≥ A . The free loop space on a manifold M is a very important object in geometry. We have seen its importance for geodesics. It is also important, for instance, for spaces of immersions. Denote by Imm(S1 , M) the space of immersions of S1 into a manifold M. There is a map from Imm(S1 , M) to the free loop space L(SM) of M’s tangent sphere bundle SM which associates to an immersion the loop of unit tangent vectors. The Hirsch–Smale theorem (see for instance Smale [239]) says that, for smooth manifolds P and Q, Imm(P, Q) has the weak homotopy type of the space of bundle maps from TP to TQ that are linear and injective on the fibers. Since the tangent bundle of S1 is trivial, the space of bundle maps from TS1 to TM has the homotopy type of the free loop space on M. Hence, after re-parametrizing to have unit speed curves, Imm(S1 , M) is weakly equivalent to L(SM).

9.5

9.5.2


The Bousfield–Peterson–Smith model

The space Map(X, Y) is in general not path connected. In [39], A. K. Bousfield, C. Peterson and L. Smith constructed a cdga that is a model for Map(X, Y) in the sense that its geometric realization has the rational homotopy type of Map(X, Y). This approach has been developed by E. Brown and R. Szczarba in [43]. In fact, if Y is nilpotent of finite type and X is of finite type with finite cohomology, then the component Map(X, Y, f ) of the mapping space Map(X, Y) containing f : X → Y is a nilpotent space of finite type. Therefore, it has a minimal model, and we describe this model now. The functor Map(X, −) is the right adjoint to the product functor X × −: Map(X × Z, Y) ∼ = Map(Z, Map(X, Y)) . The correspondence associates to a map g : Z → Map(X, Y) the map g : X× Z → Y obtained as the composition ev ◦ (1 × g), X×Z

1×g

/ X × Map(X, Y)

ev

/ Y.

Here ev denotes the evaluation map X × Map(X, Y) → Y. Let CDGA be the category of cdga’s over Q. If A = (A, d) is a fixed cdga of finite type, we consider the tensor product functor A ⊗ − : (B, d) → (A, d) ⊗ (B, d) of CDGA into itself. Since the category CDGA modelizes topology in a contravariant way, a good model for the function space is given by a left adjoint to the tensor product. Functors of this type were first systematically studied by Jean Lannes. Their utilization for models of mapping spaces is due to A.K. Bousfield, C. Peterson and L. Smith [39] (see also [232] for a coalgebra version). Proposition 9.39 Let A = (A, d) be a fixed cdga of finite type. The tensor product functor A ⊗− : (B, d) → (A, d)⊗(B, d) of CDGA into itself admits a left adjoint functor A˜ . Proof Let (∧V, d) be a minimal algebra. We consider, as in the Haefliger model, the morphism of commutative algebras θ : ∧ V → A ⊗ ∧(A∨ ⊗ V) . There is then a differential D on ∧(A∨ ⊗V) such that θ : (∧V, d) → (A, d)⊗ (∧(A∨ ⊗ V), D) is a morphism of cdga’s. We define

A˜ (∧V, d) = (∧(A∨ ⊗ V), D) , and note that this construction extends to any cdga (B, d), making A˜ into a functor A˜ : CDGA → CDGA (see [39] for details).

371

372


The adjunction process works as follows.Let ϕ : (∧V, d) → (A, d) ⊗ (B, d) be a map of cdga’s and write ϕ(v) = i ai ⊗ ϕi (v). We associate to ϕ the map ϕ : (∧(A∨ ⊗ V), D) → (B, d) defined by ϕ(a i ⊗ v) = ϕi (v).

Now we describe the process that gives the model of a connected component of function spaces. We first define the maximal connected component of an augmented cdga. Let (A, d) be a cdga and let ε : (A, d) → Q be a morphism. We consider the ideal I generated by A 2m. Then (∧(c1 , . . . , cm )/I, 0) is a model for BU(m)(2m) , and we have the next result. Proposition 9.49 The cdga (∧(c1 , . . . , cm )/I ⊗ ∧(h1 , . . . , hm ), d) = (∧(cj )/I, 0) ⊗∧(cj ) (∧(cj , hj ), d) is a model for EU(m)(2m) . The cohomology of this algebra is well known and a basis for it has been described by Vey [106]. It follows also from [106] that EU(m)(2m) has the rational homotopy type of a wedge of spheres. For instance EU(1)(2) has the rational homotopy type of S3 and EU(2)(4) has the rational homotopy type of S5 ∨ S5 ∨ S7 ∨ S8 ∨ S8 . From Theorem 9.38, we then deduce Theorem 9.50 ([260]) Let M be a compact connected nilpotent manifold of dimension m ≥ 2, whose Pontryagin classes are zero. Then there is A > 1 such that for k large enough, ki=0 dim H i (C ∗ (LM )) ≥ Ak .

9.6

Iterated integrals

While all of this section can be written in the framework of simplicial sets (see [59], [127] or [248]), for the sake of simplicity, we will stay within the world of manifolds to describe the various ideas introduced by K.T. Chen.

9.6.1

Definition of iterated integrals

Let M be a smooth manifold. Throughout this book, we have been motivated by the same recurrent theme: find homotopical information about a smooth manifold M from its algebra of differential forms. One of the simplest and well-known ways of doing this relieson differential 1-forms. If ω ∈ A1DR (M) is closed, the map α ∈ M[0,1] →

α

ω respects homotopy

classes of smooth loops. In a series of papers, K.T. Chen extended this fact to p-forms and to spaces more general than manifolds, called differential spaces. Since we will be interested only in path spaces on manifolds, we will not use this more general notion. The paper [59] gives a general overview of his theory.

9.6

Iterated integrals

The first objects introduced by Chen that we encounter are the iterated integrals. To introduce them, let’s begin with a very particular situation. Let ω1 and ω2 be two 1-forms on a smooth manifold M. If α : [0, 1] → M is a piecewise smooth path on M, for 0 ≤ t1 ≤ t2 ≤ 1, we define

ω1 ω2

t2

where 0

α

1 t2

=

0

0

ω1 (α(t1 )) dα(t1 ) ω2 (α(t2 )) dα(t2 ),

ω1 (α(t1 )) dα(t1 ) =

0

t2

dα(t1 ) ω1 (α(t1 )) dt1

dt1 .

ω1 ω2 and called an iterated integral. In

This association is denoted by

the particular case above, this is a function on the space Pfree (M) of (free) piecewise smooth paths on M. Note that Pfree (M) has the homotopy type of the space M[0,1] of free paths, thus the homotopy type of M. We can also see another way to define this function. It is more sophisticated, but it will lead us to the general case. Let k = {(t1 , t2 , . . . , tk ) | 0 ≤ t1 ≤ t2 ≤ . . . tk ≤ 1} be the euclidian simplex and let pi : k → [0, 1] be the canonical projection for i = 1, . . . , k. We denote by ω˜ i the 1-form on 2 which is the pullback of ωi along α ◦ pi . We have = ω˜ 1 ∧ ω˜ 2 . ω1 ω2 α

2

This iterated integral is nothing more than an integral on 2 and the qualification iterated comes from the manner in which we compute the integral. In the framework we are going to describe, iterated integrals are not only functions on Pfree (M), but more generally forms on Pfree (M). Since Pfree (M) is not a classical smooth manifold, we first have to give a meaning to this notion and that is where Chen’s notion of differentiable space comes in. As we mentioned above, our situation is sufficiently simple that we can avoid the introduction of this notion. Recall that a form on a manifold is defined by its trace on any trivializing open set, so the next definitions come naturally. Definition 9.51 Let M and N be two smooth manifolds. A smooth map α : N → Pfree (M) is a continuous map such that there exists a partition of [0, 1], 0 = t0 < t1 < . . . < tr = 1, for which the restrictions α : N × [ti , ti+1 ] → M of the adjoint map are smooth. The

377

378


charts of Pfree (M) are the smooth maps α where N is an open subset U of an Rn . Definition 9.52 Let M be a smooth manifold. A differential p-form on Pfree (M) is a correspondence which associates to any chart

p α : U → Pfree (M) an element ωα ∈ ADR (U), such that ωα◦ = ∗ (ωα ) for p

any smooth map : U → U. We let ADR (Pfree (M)) denote the space of differential p-forms on M and take ADR (Pfree (M)) =

/

p

ADR (Pfree (M)).

p≥0

Observe that ADR (Pfree (M)) has a cdga structure with respect to the obvious differential and the product of forms. Now, we are ready for the definition of iterated integrals as forms on Pfree (M). Definition 9.53 Let ω1 , . . . , ωk be forms on a smooth manifold M with qi ωi ∈ ADR (M). The iterated integral ω1 . . . ωk is a ((q1 + · · · qk ) − k)-form on Pfree (M) defined as follows. Let α : U → Pfree (M) be a chart with adjoint α : U × [0, 1] → M. Decompose the differential form α∗ (ωi ) on U × [0, 1] into α∗ (ωi )(x, t) = dt ∧ ωi (x, t) + ωi (x, t), where ωi and ωi do not contain the factor dt, and set

ω1 . . . ωk

α

=

k

ω1 (x, t1 ) ∧ · · · ∧ ωk (x, tk ) dt1 . . . dtk .

Observe that the form ωi is the interior product of the form α∗ (ωi ) with ∂ the vector field . Let ω˜ i,α be the pullback of the form ωi along the compo∂t sition U × k be expressed as:

idU ×pi

α

/ U × [0, 1]

ω1 . . . ωk

α

/ M . The iterated integral can also

. =

k

ω˜ 1,α ∧ · · · ∧ ω˜ k,α ,

. where

is the integral along the fiber in the trivial bundle U × k → U.

9.6

9.6.2

Iterated integrals

The cdga of iterated integrals

Let Chen Pfree (M) be the sub-vector space of forms on Pfree (M) generated by the ∗ π0 (ω0 ) ∧ ω1 . . . ωk ∧ π1∗ (ωk+1 ), where • ωi ∈ ADR (M), for i = 0, . . . , k + 1; •

ω1 . . . ωk is the iterated integral of Definition 9.53;

Pfree (M) → M are the evaluations α → α(0), α → α(1) respectively.

• π0 , π 1 :

Theorem 9.54 The complex Chen Pfree (M) is a differential graded subalgebra of ADR (Pfree (M)). This comes from Lemma 9.56 and Lemma 9.58 below which show the stability of Chen Pfree (M) under the differential and the product of forms. Before stating these lemmas, we mention an alternative presentation of the elements of Chen Pfree (M). Remark 9.55 The forms π0∗ (ω0 ) ∧ ω1 . . . ωk ∧ π1∗ (ωk+1 ) on Pfree (M) can also be presented as integrals along a fiber, see [105] or [191]. In the diagram,

Pfree (M) × k p

k ev

/ Mk+2

Pfree (M) k is the evaluation defined by the map ev k (τ , (t1 , . . . , tk )) = (τ (0), τ (t1 ), . . . , τ (tk ), τ (1)). ev Let (ωi )0≤i≤k+1 be (k + 2) forms on M and note that . ∗ ∗ k (ω0 ∧ · · · ∧ ωk+1 ) = π0 (ω0 ) ∧ ω1 . . . ωk ∧ π1∗ (ωk+1 ), ev . where

is the integral along the fiber in the trivial bundle p.

379

380


Lemma 9.56 (Coboundary of an iterated integral; [59]) qi Let ωi ∈ ADR (M) for i = 1, . . . , k. The following formula holds: d

k q1 +···+qi−1 −i (−1) ω1 . . . ωk = ω1 . . . dωi . . . ωk i=1

−

k−1

(−1)q1 +···+qi −i

ω1 . . . (ωi ∧ ωi+1 ) . . . ωk − π0∗ (ω1 ) ∧

i=1 q1 +···+qk−1 −k+1

+ (−1)

ω2 . . . ωk

ω1 . . . ωk−1 ∧ π1∗ (ωk ).

Proof First, let’s consider this formula (up to sign) and explain the origin of the terms contained in it. Recall that there is a Stokes theorem for integration along a fiber (see [112, page 311]): . . . d ω= dω ± ω. k

k

∂k

Let’s understand the last term in the particular case of 2 . The boundary ∂2 of 2 = {(t1 , t2 ) | 0 ≤ t1 ≤ t2 ≤ 1} has three faces: • the face t1 = 0 which gives the term π ∗ (ω1 ) ∧ ω2 ; 0 • the face t2 = 1 which gives the term ω1 ∧ π ∗ (ω2 ); 1 • the face t1 = t2 which gives the term ω1 ∧ ω2 . (Recall that the pull back of two forms on M along the diagonal M → M × M is the wedge product of the two forms, [112, page 209].) Now, let’s explain the signs that appear in the expression of the coboundary. Consider the case of the iterated integral of one form ω ∈ ADR (M). Let α : U → Pfree (M) be a chart. As in Definition 9.53, we decompose α∗ (ω) as α∗ (ω)(x, t) = dt ∧ ω (x, t) + ω (x, t) and observe that α∗ (dω) = −dt ∧ dω + dt ∧

∂ω ∂ω + dxj ∧ . ∂t ∂xj j

This gives:

dω α

1 ∂ω (x, t) dt dω dt + ∂t 0 0 = − d ω + ω (x, 1) − ω (x, 0) α = − d ω + π1∗ (ω)(α) − π0∗ (ω)(α).

=−

1

α

9.6

Iterated integrals

The general formula follows by induction (see [59, Proposition 1.5.2]).

It is time to recall certain conventions and definitions concerning permutation of graded objects. First, let A be the free commutative graded algebra generated by elements (ωi )1≤i≤n of respective degree qi . For any permutation σ of the set {1, 2, . . . , n}, we denote by ±σ ,(qi ) the element of {−1, 1} such that ω1 ∧ · · · ∧ ωn = ±σ ,(qi ) ωσ (1) ∧ · · · ∧ ωσ (n) . Definition 9.57 Let k and l be non-negative integers. A shuffle of type (k, l) is a permutation σ of the set 1, 2, . . . , k + l such that: σ −1 (1) < σ −1 (2) < · · · < σ −1 (k) and σ −1 (k + 1) < σ −1 (k + 2) < · · · < σ −1 (k + l). We denote by S h(k, l) the set of shuffles of type (k, l). q

i (M) for Lemma 9.58 (Product of iterated integrals; [59]) Let ωi ∈ ADR i = 1, . . . , k + l. The following product formula holds: ±σ ,(qi ) ωσ (1) ωσ (2) . . . ωσ (k+l) . ω1 . . . ωk ∧ ωk+1 . . . ωk+l =

σ ∈S h(k,l)

Proof The two terms on the left-hand side correspond to integration over k and l respectively. The right side is an integration over k+l . The result comes from the classical triangulation of a product of two simplices: k × l = (tσ (1) , . . . , tσ (k+l) )|0 ≤ t1 ≤ . . . ≤ tk ≤ 1, σ ∈S h(k,l)

0 ≤ tk+1 ≤ . . . ≤ tk+l ≤ 1 .

9.6.3

Iterated integrals and the double bar construction

We now want to compare the double bar construction of Subsection 5.9 with the complex of iterated integrals Chen Pfree (M). Let M be a connected manifold with basepoint ∗ ∈ M. We denote by (A, d) (or A if there is no ambiguity) the differential algebra ADR (M) and define an augmentation ε : A → R by f → f (∗) if f ∈ A0 and ε(ω) = 0 if ω ∈ A>0 .

381

382


q First recall that B(A, A, A) = (A ⊗ (sA ⊗ · · · ⊗ sA) ⊗ A, D) with sA = A

q+1

and the differential D as in Definition 5.51. Consider the map: B(A, A, A)

ω0 ⊗ [ω1 | . . . |ωk ] ⊗ ωk+1

ψ

/ ADR (P (M)) free / π ∗ (ω0 ) ∧ 0

ω1 . . . ωk ∧ π1∗ (ωk+1 )

Lemma 9.56 shows that ψ is compatible with the differentials. In the case A = ADR (M), the bar construction on the left-hand side can have elements of negative degree. The classical procedure for avoiding this is to replace A by an equivalent augmented complex without elements of degree 0 in the augmentation ideal. Chen uses a normalization of the bar construction that we present now (see [58, page 22] or [105, page 353] for details). If f ∈ A0 , set Si (f )[ω0 | . . . |ωk ] = [ω0 | . . . |ωi−1 |f |ωi | . . . |ωk ]. Define Degen(A) as the subspace of B(A, A, A) generated by the Si (f ) and the [D, Si (f )]. Chen proves: Proposition 9.59 Let A = ADR (M). The space Degen(A) is stable under the differential D and is acyclic if M is connected. Thus, the quotient N(A, A, A) = B(A, A, A)/Degen(A) has the same cohomology as B(A, A, A). Moreover, since ADR (Pfree (M)) has no elements of negative degree, the map ψ above induces a map (which we still denote by) ψ : N(A, A, A) → 0 ADR (Pfree (M)). Finally observe that B(A, A, A) = N(A, A, A) if A = 0. We would also like to consider the other bar constructions presented in Subsection 5.9 and to relate them to certain complexes of iterated integrals. For this, we introduce: • the subspace P (M) of P (M) consisting of the paths with origin ∗ ∈ M; ∗,a free • the subspace P (M) of P (M) consisting of the based loops on (M, ∗); ∗,∗ free • the cdga’s ADR (P (M)) and ADR (P (M)) as in Definition 9.52. ∗,a

∗,∗

Observe that • • •

Pfree (M) is of the homotopy type of M[0,1] , therefore of M; P∗,a (M) is of the homotopy type of P(M), therefore is contractible; P∗,∗ (M) is of the homotopy type of the loop space (M).

For the proof of the next result, we refer the reader to the paper of Chen [59] (see also [105] with a slightly different sign convention).

9.6

Iterated integrals

Theorem 9.60 (Chen, [59]) Let M be a simply connected manifold and A = ADR (M). The following maps of complexes are quasi-isomorphisms: 1. ψ : N(A, A, A) → ADR (Pfree (M)), ω0 ⊗ [ω1 | . . . |ωk ] ⊗ ωk+1 → π0∗ (ω0 ) ∧ ω1 . . . ωk ∧ π1∗ (ωk+1 ); 2. ψ : N(A, A, R) → ADR (P∗,a (M)), ω0 ⊗ [ω1 | . . . |ωk ] → π0∗ (ω0 ) ∧ ω1 . . . ωk ; 3. ψ : N(R, A, R) → ADR (P∗,∗ (M)), [ω1 | . . . |ωk ] → ω1 . . . ωk . As a consequence, the images of these maps give the cohomology of the corresponding spaces. In particular, recalling from Section 5.9 that B(A, A, A) is quasi-isomorphic to A, we have Corollary 9.61 If M is a simply connected manifold, then the two cdga’s, ADR (Pfree (M)) and Chen Pfree (M), have the same minimal model as M. Let’s focus on the last item of Theorem 9.60 which gives a model for the loop space (M). Let Chen P∗,∗ (M) be the subspace of ADR (P∗,∗ (M)) generated by the iterated integrals ω1 . . . ωk , where ωi ∈ ADR (M). Remark 9.62 This complex can also be expressed using a presentation similar to Remark 9.55. In the diagram

P∗,∗ (M) × k p

evk

/ Mk

P∗,∗ (M) the map p is the canonical projection and evk is the evaluation defined by evk (τ , (t1 , . . . , tk )) = (τ (t1 ), . . . , τ (tk )). Definition 9.53 coincides with: . ω1 . . . ωk = ev∗k (ω1 ∧ · · · ∧ ωk ), . where

is the integral along the fiber in the trivial bundle p.

Finally, observe that in P∗,∗ (M) the maps π0 ◦α and π1 ◦α are constant, so that π0∗ (ω) = π1∗ (ω) = 0 for any ω ∈ A≥1 (M). This presentation is nothing other than Remark 9.55 restricted to P∗,∗ (M). In [59, Theorem 4.1.1], Chen proves that [ω1 | . . . |ωk ] → ω1 . . . ωk induces an isomorphism of Hopf algebras between N(R, A, R) and

383

384


Chen P∗,∗ (M). The product is the shuffle product of Lemma 9.58 and the diagonal is given by: k ω1 . . . ωk = ω1 . . . ωi ⊗ ωi+1 . . . ωk . i=0

This and Theorem 5.52 directly imply the next statement. Corollary 9.63 [(59]) Let M be asimply connected manifold with loop space (M). The map [ω1 | . . . |ωr ] → ω1 . . . ωr induces an isomorphism of Hopf algebras between N(R, A, R) and Chen P∗,∗ (M). This gives an isomorphism of Hopf algebras between the cohomology of the loop space (M) and the cohomology of the complex of iterated integrals:

H ∗ ((M); R) ∼ = H ∗ Chen P∗,∗ (M) .

9.6.4

Iterated integrals, the Hochschild complex and the free loop space

We would like now to study the cohomology of the free loop space with iterated integrals. Denote by Pfree loop (M) the subspace of Pfree (M) consisting of the free loops on M. Observe that Pfree loop (M) has the homotopy type of the free loop space LM. Let Chen Pfree loop (M) denote the subspace of forms on Pfree loop (M) generated by the ∗ π0 (ω0 ) ∧ ω1 . . . ωk . We leave to the reader the descriptions of these forms as integrals along a fiber as in Remarks 9.55 and 9.62. As a complement to Section 5.9, note that the shuffle product on B(ADR (M)) gives a cohomology algebra isomorphism with H ∗ (LM; R). As in Subsection 9.6.3, we observe that B(ADR (M)) has elements of negative degree. We kill them by quotienting by a differential ideal Degen(A) which is acyclic when M is connected. Theorem 9.64 [(59, Theorem 4.2.1], [105, Proposition 4.1]) Let M be a simply connected manifold. The map

B(ADR (M)) ω0 ⊗ [ω1 | . . . |ωk ]

ϕ

/ ADR (Pfree loop (M)) / π ∗ ω0 ∧ 0

ω1 . . . ωk

induces an isomorphism of algebras between B(ADR (M)) and Chen Pfree loop (M). Therefore, there is an isomorphism of algebras,

H ∗ Chen Pfree loop (M) ∼ = H ∗ (LM; R).

9.6 Table 9.1

Iterated integrals

Chen’s iterated integrals.

Iterated integrals

Bar constructions

Spaces

Chen Pfree (M)

B(A, A, A)

M[0,1]

Chen P∗, a (M)

B(A, A, R)

P(M)

Chen P∗,∗ (M)

B(R, A, R)

(M)

Chen Pfree loop (M)

B(A)

LM

Table 9.1 gives a summary of the correspondences between the various complexes of iterated integrals, bar constructions and particular path spaces.

9.6.5

Formal homology connection and holonomy

In this paragraph, we replace Chen P∗,∗ (M) by a differential algebra, free as an algebra, whose homology is isomorphic to the Pontryagin algebra of the loop space, (M), of a simply connected manifold M with finite Betti numbers. Denote by V the subspace s−1 H+ (M; R); that is, Vn = Hn+1 (M; R) for all n ≥ 1 and V0 = 0. The free graded Lie algebra on V, denoted by L(V), is a sub-vector space of the free tensor algebra T(V), and if T(V) is endowed with the Lie algebra structure coming from the commutator bracket, then L(V) is the sub-Lie algebra generated by V, see [87, Page 289]. A Lie derivation on L(V) is a map ∂ : L(V)n → L(V)n−1 such that ∂[l, l ] = [∂l, l ] + (−1)|l| [l, ∂l ] for all homogeneous elements l and l of L(V). Finally, we introduce the completion of the tensor product as: L(V) = ADR (M)⊗

/ n

n−p

(ADR (M) ⊗ L(V)p ).

p

This is the set of all formal sums i ωi ⊗ li such that, for each integer n, there are only a finite number of terms ωi ⊗ li such that |ωi | + |li | = n. L(V) is a graded Lie algebra with a bracket defined Recall that ADR (M)⊗ on homogeneous elements by [a ⊗ b, a ⊗ b ] = (−1)|a | |b| aa ⊗ [b, b ]. The map d is extended to the tensor product by setting d(a ⊗ b) = da ⊗ b on homogeneous elements. Similarly, if ∂ is a Lie derivation on L(V), we define ∂(a ⊗ b) = (−1)|a| a ⊗ ∂b. Definition 9.65 Let M be a compact simply connected manifold. Set V = s−1 H∗ (M; R) and fix a homogeneous basis (Xi )i∈I of V. A formal homology

385

386


connection is a pair (w, ∂) with w= ωi ⊗ Xi + ωi1 i2 ⊗Xi1 Xi2 + · · · I

(i1 ,i2 )∈I×I

L(V) ⊂ ADR (M)⊗ T(V) ∈ ADR (M)⊗ and ∂ a Lie derivation on L(V) such that: • the ωi are closed elements of ADR (M); • the family ([ωi ])i∈I ∈ H ∗ (M; R) is the dual basis of (sXi )i∈I ; • w is of degree −1, which means |ωi ...i | = 1 + r (|ωi | − 1); j=1 1 r j • d and ∂ satisfy the Maurer–Cartan equation dw = ∂w + 1 [w, w]. 2

We give below an equivalent definition in the context of maps instead of tensor products. First, let’s consider an easy example. Example 9.66 Consider the space CP(2) with its real cohomology denoted by H ∗ and real homology denoted by H∗ . We choose a basis X1 ∈ V1 = s−1 H2 , X2 ∈ V3 = s−1 H4 . Let ω be the Kähler form in A2DR (CP(2)), and let ([ω], [ω2 ]) be a dual basis of (X1 , X2 ). We set ∂X1 = 0, ∂X3 = 12 [X1 , X1 ] and extend it to L(X1 , X2 ) as a Lie derivation. An easy computation shows that w = ωX1 + ω2 X2 is a formal homology connection on CP(2). Remark 9.67 A formal homology connection can be viewed as a differential form with values in a Lie algebra, the free Lie algebra L(V). If we consider the classical case of a form ω ∈ A1 (M; g) where g is the Lie algebra associated to a Lie group G, then we see that ω can be extended to a morphism of graded algebras ω : C ∗ (g) → ADR (M) where C ∗ (g) is the classical cochain algebra on g. This map ω is compatible with the differentials if and only if dω − 12 [ω, ω] = 0. The analogue of the classical situation exists for the formal homology connection of Chen as shown by the next result. Theorem 9.68 Let (ω, ∂) be a formal homology connection on a simply connected compact manifold M. We choose a basis (xJ ) of L(V) and decompose L(V). Let ω as ω = J ωJ ⊗ xJ ∈ ADR (M)⊗ ω : C ∗ (L(V), ∂) → ADR (M) be defined by ω (w) = J ±ωJ w, xJ , where C ∗ (L(V), ∂) = (∧W, d) is the cochain algebra introduced in Subsection 2.6.2, w ∈ W and , is the duality between W •+1 and L(V)• . Then the following are true: 1. ∂ 2 = 0; 2. ω is compatible with the differentials; 3. ω induces an isomorphism in cohomology.

9.6

Iterated integrals

For the proof, we refer the reader to [59] or [248, Theorem IV. 2.(1)] and observe, as a direct consequence, that a formal homology connection gives a Sullivan model. Corollary 9.69 Let (ω, ∂) be a formal homology connection on a simply connected compact manifold M. The map ω defined in the statement of Theorem 9.68 is a Sullivan model of M. We have not introduced differential Lie algebra models here, yet we note that Theorem 9.68 contains within it the fact that (L(V), ∂) is such a model. In fact, if we let (T(V), ∂) denote the associative tensor algebra obtained as the enveloping algebra of (L(V), ∂), then we can use a holonomy map $ : C∗ ((M)) → (T(V), ∂) constructed by Chen to give an isomorphism of algebras between the Pontryagin algebra H∗ ((M); R) and H∗ (T(V), ∂), see [59, Section 3.4] or [248, Proposition IV.3.(1)] for a different approach to the construction of the map $.

9.6.6

A topological application

In Subsection 4.6.5, we considered the question of which symplectic manifolds are nilpotent spaces (and thus are amenable to minimal model techniques, see Proposition 4.100). Iterated integrals can be used to approach this question (but not answer it completely yet). We will only state the result and refer the reader to the original paper of Chen [57]. Theorem 9.70 Let Mn be a connected manifold and suppose ω1 , . . . , ωm are closed 1-forms on M which satisfy the following conditions: • ωi ∧ ωj = 0 for i, j = 1, . . . , m; • the cohomology classes [ω1 ], . . . , [ωm ] are linearly independent.

Then π1 (M) contains a free subgroup G of rank m. Moreover, if ω is a closed q-form on M such that • ωi ∧ ω = 0 for i = 1, . . . , m; • for some α ∈ πq (M), the integral α ω = 0;

then {β · α}β∈G is a basis of a free abelian subgroup of πq (M). This result allows us to see the action of π1 (M) on πq (M) in terms of differential forms. Indeed, we can also see Proposition 4.100 from this viewpoint. Can we use Theorem 9.70 or an extension of it to understand nilpotency in the context of symplectic manifolds?

387

388


9.7

Cohomological conjectures

Throughout the book we have indicated a series of conjectures related to geometrical problems. Our purpose in this section is to recall them in a more general setting.

9.7.1

The toral rank conjecture

The toral rank conjecture states that if a torus T r acts almost freely on a compact manifold M, then dim H ∗ (M; Q) ≥ 2r . With the help of Borel fibrations, it can be reformulated in terms of fibrations in the following form: Conjecture 9.71 If F → E → B is a quasi-nilpotent fibration (e.g. if π1 (B) = 0) where F is a torus T r and the cohomology of B is finite dimensional, then dim H ∗ (E; Q) ≥ 2r . This leads to the more general Problem 9.72 Let F → E → B be a quasi-nilpotent fibration, where F is a rationally elliptic space and the cohomology of B is finite dimensional. Is it true that dim H ∗ (E; Q) ≥ dim H ∗ (F; Q)?

9.7.2

The Halperin conjecture

The Halperin conjecture is also related to fibrations. Let’s recall it and note the strong relation with the conjecture above, see also Section 4.2. Conjecture 9.73 If F → E → B is a quasi-nilpotent fibration, where F is a rationally elliptic space with χ(F) > 0, then the Serre spectral sequence degenerates at the E2 term: that is, there is an isomorphism of H ∗ (B; Q)modules, H ∗ (E; Q) ∼ = H ∗ (F; Q) ⊗ H ∗ (B; Q). A positive answer to the Halperin conjecture would give a positive answer to Open Problem 9.72 in the case χ(F) > 0. There are equivalent forms of the Halperin Conjecture: Let X be an elliptic space with χ (X) > 0. Then the following conditions are equivalent (see Proposition 4.40): 1. The Halperin conjecture is true for X. 2. H ∗ (X; Q) has no non–zero derivation of negative degree. 3. The space of self-homotopy equivalences of X, aut X, has the rational homotopy type of a product of odd dimensional spheres.

9.7


The relation between the second and the third characterization follows from the following fact. The commutator of two homotopy selfequivalences gives to π∗ ( aut X) ⊗ Q the structure of a graded Lie algebra, and there is a morphism of graded Lie algebras πq ( aut X) ⊗ Q → Der H −q (X; Q) that is surjective when X is formal. The derivation viewpoint has a geometric meaning. As we saw in Chapter 6, Belegradek and Kapovitch proved that when M is a compact manifold and H ∗ (M; Q) has no nonzero negative degree derivation, then, for every torus T and every vector bundle ξ over C × T with secE(ξ ) ≥ 0, there is a finite covering p : T → T such that (p∗ × id)(ξ ) = ξC × T where ξC is a vector bundle over C. Since bundles do not generally have this property, this then gives an obstruction to positive sectional curvature on total spaces of bundles. In fact, however, Corollary 6.41 is true even when only the cohomology derivations which descend from derivations of the minimal model vanish. The question is, how can these be recognized? Let MC denote the minimal model of C. There is an obvious graded Lie algebra homomorphism ρ : H ∗ (Der(MC )) → Der(H ∗ (C)), but knowledge of Im(ρ) is scant. Of course, if C is formal, then ρ is surjective, but otherwise, we know little. Problem 9.74 Determine Im(ρ) for various classes of spaces C. In particular, determine when Im(ρ) = 0. Denote by Baut1 (C) the classifying space for the monoid of selfequivalences that are homotopic to identity. This space classifies fibrations with simply-connected bases and fibers of the homotopy type of C. The differential graded Lie algebra (Der(MC ), dC ) is a Lie algebra model for Baut1 (C). This provides an interesting connection between curvature properties and abstract homotopy theory. Furthermore, while it was proved in [237] that homogeneous spaces of maximal rank have no negative degree cohomology derivations (and so satisfy the Halperin Conjecture), the following is as yet unknown. Problem 9.75 Show that a biquotient G//H has no negative degree cohomology derivations if rk(G) = rk(H).

9.7.3

The Bott conjecture

Recall from Section 6.4 that we have the following conjecture of Bott. Conjecture 9.76 If M is a closed manifold with secM ≥ 0, then M is rationally elliptic.

389

390


Of course, the power of the conjecture (if true) is in the contrapositive statement. Since most manifolds are rationally hyperbolic, this would say that most manifolds cannot carry metrics of non-negative sectional curvature. Also, consider Bott’s conjecture in relation to Open Problem 9.72 above. Problem 9.77 Let F → E → B be a quasi-nilpotent fibration, where F has secF ≥ 0 and the cohomology of B is finite dimensional. Is it true that dim H ∗ (E; Q) ≥ dim H ∗ (F; Q)? Conjecture 9.76 is related to many others as we noted in Section 6.4. At the present time, there seems to be no good approach. Even the results of Chapter 6 lack a certain coherence, at least in terms of their relations to minimal models and the viewpoint of rational homotopy theory. Of course, there is also Paternain’s result saying that certain geodesic flows only occur on rationally elliptic manifolds. So we might fairly pose the Problem 9.78 Make the connection between geometric constraints on manifolds and algebraic constraints on minimal models systematic. The reader may well ask what our precise meaning is here. Evidently, that is part of the problem.

9.7.4

The Gromov conjecture on LM

Let M be a rationally hyperbolic manifold. Gromov conjectured that the 1 cohomology of the free loop space LM = MS has exponential growth. On the other hand, Vigué-Poirrier proved that the Betti numbers of the function space from a compact manifold into a wedge of (at least two) spheres has exponential growth. She conjectured [260] that, for any fibration p : F → E → B, where F has the rational homotopy type of a wedge of spheres, the cohomology of the space of sections of p has exponential growth. Both conjectures merge into a common problem concerning the homology of function spaces. We can state this as Problem 9.79 Let F → E → B be a quasi-nilpotent fibration where F is rationally hyperbolic and B is a compact manifold. We suppose moreover dim B = n and F is (n + 1)-connected. Do the Betti numbers of the space of sections have exponential growth?

9.7.5

The Lalonde–McDuff question

F. Lalonde and D. McDuff have asked the following Question 9.80 (Lalonde–McDuff question) Is every Hamiltonian bundle TNCZ?

9.7


Recall that this says that if F → E → B is a symplectic fiber bundle where the structure group of the bundle reduces to the Hamiltonian diffeomorphisms, then the fibration is TNCZ. Of course, we know that all bundles are TNCZ when the fibers are simply connected Kälher manifolds or maximal rank homogeneous spaces. We have also seen in Theorem 7.75 that the Lalonde–McDuff question is answered affirmatively for nilmanifolds. Problem 9.81 Use models to examine the Lalonde–McDuff question in other cases of interest.

391

Appendix A De Rham forms

In this appendix we give basic definitions and results about differential (also known as de Rham) forms. General references for the whole appendix are [107], [112] and [262].

A.1

Differential forms

We shall begin by considering Euclidean space Rn . Of course, Rn is a smooth manifold with the topology naturally given by the standard vector space dot product structure. We can define vector fields on Rn by choosing a vector space basis {e1 , . . . , en } and defining the vector field Ej by Ej (x1 , . . . , xn ) = ej at the point (x1 , . . . , xn ) ∈ Rn . In other words, Ej is simply ej placed at every point of Rn . That such a definition gives a smooth vector field is due to the fact that the coordinates are global. Any vector field on Rn , with V : Rn → Rn × Rn , now has the form V(x1 , . . . , xn ) = (x1 , . . . , xn , V) 1 , . . . , xn ) = V(x

n

aj (x1 , . . . , xn ) ej ,

j=1

where the coefficients aj are now smooth functions on Rn . It is apparent that we are thinking of Rn × Rn here as pairs (x, vx ), where vx is an nvector starting from x. This is the first instance of what is called the tangent bundle of a manifold. Here, it is very simple because the topology of Rn is simple, but it can be more complicated as we shall see. In order to be able to understand these more complicated situations, let’s look a little more closely at the Rn case now. Let x ∈ Rn . Consider a smooth curve β : (− , ) → Rn defined on a small interval (which for convenience we take to be (− , )) with β(0) = x. Of course, we now have n functions βi : (− , ) → R defined as the coordinate functions of β, β(t) = (β1 (t), . . . , βn (t)).

A.1

Differential forms

The tangent vector of β at 0 is then computed coordinatewise, β (0) = (β1 (0), . . . , βn (0)), and we define the collection of all vectors arising as tangent vectors to curves through x to be the tangent space, Tx (Rn ), to Rn at x. It is clear that each vector through x is the tangent vector to a curve, and therefore Tx (Rn ) identifies to the vector space of vectors at x. The tangent bundle of Rn is defined to be T(Rn ) = {(x, vx ) | x ∈ Rn and vx ∈ Tx (Rn )}. If f : Rn → Rk is a smooth map, then it induces a map Df : T(Rn ) → T(Rk ) of tangent bundles which is a linear map on the tangent space at each point. For f (x) = y, the map Df is defined as follows. Let (x, vx ) ∈ T(Rn ) and suppose β is a curve in Rn that defines vx as above. Then the composition α = f ◦ β is a smooth curve in Rk with α(0) = f ◦ β(0) = f (x) = y. We then define Df (x, vx ) = (y, α (0)). Note that, in Rn , we can always choose β(t) = x + t vx , the line through x in the vx -direction. If we are only interested in the map Df restricted to a particular tangent space Tx (Rn ), then we denote this linear map by Dx f . The vector α (0) is given by d(f (β(t)) d(f (x + t vx ) = , α (0) = t=0 dt dt t=0 where we have taken the straight line through x as β. This is exactly the definition of the usual directional derivative of vector calculus. Therefore, the map Df is just a generalization of that. Indeed, by looking at the coordinate functions αi of α, we see that Df is just the vector consisting of the directional derivatives of the αi . The directional derivative also allows us to transform functions into new functions. Namely, if V is a vector field and f : Rn → R is a smooth function, then the smooth function Vf is defined by d(f (x + t Vx ) Vf (x) = , dt t=0 the usual directional derivative in Rn in the direction of Vx (the value of V at x). If V = Ej , then it is easy to see that Vf = ∂f /∂xj , the usual partial derivative in the jth coordinate direction. Now consider the dual space Rn ∗ with dual basis {de1 , . . . , den }; that is, 0 if i = j dei (ej ) = 1 if i = j.

393

394

A : De Rham forms

We can also take the exterior algebra on this basis, (n) = (de1 , . . . , den ). This is the free graded algebra satisfying the anti-commutativity condition a ∧ b = −b ∧ a. Note that this means that dei ∧ dei = 0 for all i = 1, . . . , n. An element φ of degree p has the form φ= ai1 ···ip dei1 ∧ · · · ∧ deip

where the sum is over all np choices of p dej ’s among the n dej ’s available and the ai1 ···ip ’s are constant. Note that, by the anti-commutativity property of the exterior algebra, we can always order the ij so that i1 ≤ i2 ≤ . . . ≤ ip . The algebra (n) appears to be the dual of the exterior algebra (Rn ). Note that in order to apply an element of the exterior algebra to vectors, we need to have a definition that takes account of the anti-commutativity of the exterior algebra. For instance, if we simply said that de1 ∧ de2 (v, w) = de1 (v) · de2 (w), then we would not have de1 ∧ de2 = −de2 ∧ de1 . Just substitute e1 for v and e2 for w and get 1 and 0 for the respective calculations. So, the application of an exterior algebra element dei1 ∧ . . . ∧ deip to vectors v1 , . . . , vp is defined by (−1)sgn(σ ) dei1 (vσ (1) ) · · · deip (vσ (p) ), dei1 ∧ . . . ∧ deip (v1 , . . . , vp ) = σ ∈ p

where the sum is over all permutations σ in the permutation group p . This definition is compatible with anti-commutativity and will provide a global wedge product below. In fact, we can define 1-forms dEi dual to the vector fields Ei by requiring 0 if i = j dEi (Ej )(x1 , . . . , xn ) = 1 if i = j. Note that, at any point (x1 , . . . , xn ) ∈ Rn , we have Ei (x1 , . . . , xn ) = ei for each i = 1, . . . , n. So Ei and 1-forms dEi are defined globally and evaluate to the corresponding dual vector space basis elements at every point in Rn . Because of the obvious identifications, we may sometimes use the notation dei for global forms. Just as for vector fields, we define a p-differential form α : Rn → Rn × p α ) with (Rn ) by α(x1 , . . . , xn ) = (x1 , . . . , xn , α (x1 , . . . , xn ) = ai1 ···ip (x1 , . . . , xn ) dEi1 (x1 , . . . , xn ) ∧ · · · ∧ dEip (x1 , . . . , xn ) = ai1 ···ip (x1 , . . . , xn ) dei1 ∧ · · · ∧ deip ,

A.1

Differential forms

where the sum is taken over all multi-indices of length p and the ai1 ···ip are smooth functions on Rn . From now on, we will simply denote a form α by its second coordinate, α . A 1-form α = ai dEi may be applied to a vector field V = cj Ej to produce a smooth function on Rn as follows: α(V)(x1 , . . . , xn ) = α(x1 , . . . , xn )(V(x1 , . . . , xn )) ⎞ 3⎛ n 2 n ai (x1 , . . . , xn ) dei ⎝ cj (x1 , . . . , xn ) ej ⎠ = i=1

=

n

j=1

ak (x1 , . . . , xn ) · ck (x1 , . . . , xn ),

k=1

using the duality between dei and ej . More generally a p-form α = ai1 ···ip dEi1 ∧ · · · ∧ dEip acts on a p-tuple of vector fields V1 , . . . , Vp as follows: α(V1 , V2 , . . . , Vp )(x) =

ai1 ···ip (x) (dei1 ∧ · · · ∧ deip )((V1 )x , . . . , (Vp )x ) .

Moreover, if f : Rn → R is a smooth function, then we may define a 1-form df by df (V)(x) = Vf (x) for all vector fields V. This is our first example of what will be called exterior differentiation. The vector space formed by the forms is equipped with a multiplication, called the wedge product. Let α = ai dEi1 ∧ · · · ∧ dEip be a p-form and β = bj dEj1 ∧ · · · ∧ dEjq be a q-form. The wedge product α ∧ β is defined to be the (p + q)-form α∧β =

ai bj dEi1 ∧ · · · ∧ dEip ∧ dEj1 ∧ · · · ∧ dEjq .

i,j

If Y1 , . . . , Yp+q are vector fields, we have, (α ∧ β)(Y1 , . . . , Yp+q ) (−1)sgn(σ ) α(Yσ (1) , . . . , Yσ (p) ) · β(Yσ (p+1) , . . . , Yσ (p+q) ). = σ ∈ p+q

Here, the sum is over all permutations σ in the permutation group p+q . It is also possible to integrate n forms (as the notation dej suggests). If a form has the expression α = a dE1 ∧ . . . ∧ dEn , then its integral over some

395

396

A : De Rham forms

region W ⊆ Rn is defined to be α= · · · a(x1 , . . . , xn ) dE1 (x1 , . . . , xn ) ∧ . . . ∧ dEn (x1 , . . . , xn ) W = · · · a(x1 , . . . , xn ) de1 ∧ . . . ∧ den = · · · a(x1 , . . . , xn ) dx1 . . . dxn , where we have used the more familiar integral notation using coordinates. Note that, since dEi ∧ dEj = −dEj ∧ dEi , the integral is a signed integral depending on the order in which integrals are taken. The Rn -form ω = dE1 ∧ . . . ∧ dEn is called a volume form on Rn because if we take its integral over any particular region W, we obtain the signed volume of W. If f : Rk → Rn is a smooth function and β is a p-form on Rn , then a ∗ k new p-form f β on R is defined by “pulling back” β: namely, if β = βi dEi1 . . . dEip , then (βi ◦ f ) dfi1 ∧ . . . ∧ dfip , f ∗β = ∂fir ∗ with dfir = ∂xk dxk . If V1 , . . . , Vp are vector fields, then f β(V1 , . . . , Vp )(x) = β(Dfx ((V1 )x ), . . . , Dfx ((Vp )x )), where Dfx is the derivative map of f on tangent vectors. This now leads us to a more general situation. If Mn is a smooth n-manifold, then M is covered by charts (i.e. homeomorphisms onto open sets) ντ : Uτ → Rn such that the transition functions ντ γ = ντ νγ −1 (restricted to the appropriate images) are smooth functions from Rn to itself. The charts, in effect, parametrize M allowing calculations to be done in euclidean space. For instance, tangent spaces are defined using our definition in Rn . If β : (− , ) → M is a curve with β(0) = x ∈ Uτ ⊂ M, then α = ντ β : (− , ) → Rn is a smooth curve with a tangent vector α (0) at ν(x). We then say that Tx M consists of all such vectors. (If M is embedded in some large RN , then we can visualize the tangent space just as we do for surfaces embedded in R3 .) We then define the tangent bundle to M to be TM = {(x, vx ) | x ∈ M and vx ∈ Tx M}, provide a where Tx M is the tangent space to M at x. Local coordinates ai ei as for Rn . Of basis for Tx M, {e1 , . . . , en }, and we may write vx = course, in the particular chart, the coefficients ai are functions, changing as x changes. There is a projection p : TM → M given by p(x, vx ) = x. A vector field V on M is then a smooth mapping V : M → TM such that p V(x) = x for all x ∈ M. It should be mentioned, however, that it is not in

A.1

Differential forms

general the case that the tangent bundle of M is a product M × Rn as was true for Rn , because otherwise each manifold would support vector fields that never vanish, which is not the case for instance for S2 . When the vector field is restricted to a chart Uτ , then we can write the tangent bundle over Uτ as such a product. Charts also tell us whether M has consistent orientations of tangent spaces as we move around the manifold. Each transition function ντ γ is a mapping from an open set in Rn to another open set in Rn . Therefore, if we consider the linear map on tangent spaces Dx ντ γ : Tx Rn → Ty Rn (where y = ντ γ (x)), then we may take det(Dx ντ γ ). Since det(Dx ντ γ ) is always different from 0, its sign is the same on any connected domain of the transition map, so the question of consistent orientation reduces to looking at all possible overlaps of charts on the manifold. If charts can be chosen covering M so that det(Dx ντ γ ) > 0 for all choices of τ and γ , then M is orientable. For example, spheres, tori, odd-dimensional real projective spaces and complex projective spaces are orientable. The Klein bottle and even-dimensional real projective spaces are nonorientable. We have the following useful result. Proposition A.1 Every simply connected manifold is orientable. We may view forms by their images in the different charts. A p-form on M can be seen as a collection of p-forms on the open images of all charts in Rn obeying a compatibility relation. Namely, for a p-form on M, α = {ατ }τ , we must have ντ∗γ ατ = αγ . With this definition, we now can treat forms locally by fixing a chart (i.e. a coordinate system) and writing the local expression for the form as above for Rn . With this in mind, note that we now can define a wedge product of forms on M simply by taking wedge products in the images of charts in Rn and then forming the collection: that is, if α = {ατ }τ and β = {βτ }τ , then α ∧ β = {ατ ∧ βτ }τ . A volume form on Mn is an n-form ω that is non-zero at every point. It is not true that every manifold has a volume form. For example, for surfaces embedded in R3 , the existence of a volume form is equivalent to the existence of a nonvanishing normal vector field over the whole surface. As the example of the Möbius strip shows, such vector fields sometimes are forced to be zero at points on the surface. The Möbius strip is nonorientable and this is in fact the criterion for nonexistence of a volume form. Proposition A.2 A compact manifold M has a volume form if and only if it is orientable. In particular, by Proposition A.1, every compact simply connected manifold has a volume form. Note that a vector field on M also has a local expression in terms of chosen coordinates, so all local computations may be carried out as in Rn .

397

398

A : De Rham forms

The cotangent bundle and pth exterior algebra bundle are defined to be T ∗ M = {(x, φx ) | x ∈ M and φx ∈ Tx ∗ M}, p (T ∗ M) = {(x, φx ) | x ∈ M and φx ∈ p (Tx ∗ M)}, where Tx ∗ M is the dual vector space to Tx M and ∗ (Tx ∗ M) is the exterior algebra on it. There are obvious projections q : T ∗ M → M and q : p T ∗ M → M here as well. A p-form α on M is then seen to be a smooth mapping α : M → p (T ∗ M) such that q α(x) = x for all x ∈ M. In this way, the geometry of M is linearized. We denote the p-forms on M p by ADR (M) and the entire graded algebra of forms by ADR (M).

A.2

Operators on forms

There are several important operations that are performed on forms to create new ones illuminating certain geometric properties. Let Mn be a manifold with a vector field X and a (p + 1)-form α defined on it, then we define a p-form i(X)α by (i(X)α)(Y0 , . . . , Yp−1 ) = α(X, Y0 , . . . , Yp−1 ) , where Y0 , . . . , Yp−1 are vector fields. The operation on the (p + 1)-form α that produces the p-form i(X)α is called interior multiplication by X. Interior multiplication is an anti-derivation (or in more modern terminology, a graded algebra derivation), meaning that i(X)(α ∧ β) = (i(X)α) ∧ β + (−1)|α| α ∧ (i(X)β). If X and Y are vector fields, then their bracket [X, Y] is defined by its action on smooth functions, [X, Y]f = X(Yf ) − Y(Xf ). The bracket is obviously anti-commutative by definition. It also satisfies a Jacobi identity (see Exercise A.1), [[X, Y], Z] + [[Y, Z], X] + [[Z, X], Y] = 0, that will be important when we consider Lie groups. Here the action of a vector field on a function is slightly different from that in Rn . There may be no straight line x + t Vx on M in the definition, so we modify the definition as follows. Let X be a vector field and let f be a function on M. At x ∈ M choose a curve β : (− , ) → M such that β (0) = Xx . Then d(f (β(t)) . Xf (x) = dt t=0

A.2

Operators on forms

If X is a vector field on M, then there is an associated flow φ : R ×M → M given by fixing x ∈ M and finding the solution of the differential equation dφ(t, x) = Xφ(t,x) , dt with φ(0, x) = x. The flow is also written φt (x). For fixed x, this is a curve c(t) = φt (x) starting at x and following the vector field X in the sense that its tangent vector at c(t) is Xc(t) . For fixed t, φt : M → M is a diffeomorphism of M. Therefore, it can be used to pull back forms. To see the effect of the flow (or equivalently, the vector field X) on forms, we can find the rate of change of the pulled-back form φt ∗ α minus the original form α at the same point. The definition is φt ∗ αφ(t,x) − α(x) . t→0 t

L(X)α(x) = lim

The operation L(X) is called the Lie derivative with respect to X. We can also express L(X)α by its action on vector fields. Let α be a p-form, then

L(X)α(Y0 , . . . , Yp−1 ) p−1

= X(α(Y0 , . . . , Yp−1 )) −

α(Y0 , . . . , [X, Yj ], . . . , Yp−1 ).

j=0

The Lie derivative L(X) is a degree 0-derivation on forms:

L(X)(α ∧ β) = (L(X)α) ∧ β + α ∧ (L(X)β). In [112, page 157] it is shown that the following formula holds (see Exercise A.2): ∗

φt α − α =

t 0

φs ∗ L(X)α ds.

Then we can see that φt ∗ α = α if and only if L(X)α = 0. In this way the Lie derivative can tell us that α is invariant along trajectories of a vector X. A form α that has L(X)α = 0 for a vector field X is said to be L(X)-invariant. We defined the differential of a function f to be the 1-form df satisfying df (X) = Xf for a vector field X. The operator d is called the exterior derivative and is defined on all forms. For a (p − 1)-form α and vector fields

399

400

A : De Rham forms

Y0 , . . . , Yp−1 , the p-form dα is defined by, p−1

dα(Y0 , . . . , Yp−1 ) =

j , . . . , Yp−1 )) (−1)j Yj (α(Y0 , . . . , Y

j=0

+

i , . . . , Y j , . . . , Yp−1 ), (−1)i+j α([Yi , Yj ], Y0 , . . . , Y

i

> >· · · · c>FF > > >FFF > > FF > > FFF > > FF F · >dII FF · > · · >II FF d 4> >III FF > > IIIFF > > IIFFF > · ·> gO>OOIIIFIFF·> > · F >OOOIOIIFIFF> > OOOIOIFIF> OOIFOIF> d>2

·

·

·

·

·

· /

p

The elements of total degree n are on the line p + q = n and a differential d i joins a point of the line p + q = n to a point of the line p + q = n − 1.

B.2

Spectral sequences in cohomology

We have represented d 2 , d 3 and d 4 above. We also have: r • Er+1 p,q = Ep,q if r > max(p, q + 1); • Er+1 is always a submodule of Er and we have a sequence of inclusions p,0 p,0 p+1

E∞ p,0 = Ep,0

/ ···

/ E3 p,0

/ E2 ; p,0

• Er+1 is always a quotient of Er and we have a sequence of surjections 0,q 0,q

E20,q

/ / E3

0,q

/ / ···

/ / Eq+2 = E∞ . 0,q 0,q

Definition B.3 In a first quadrant spectral sequence, the transgression is the relation from E2q,0 in E20,q−1 given by the diagram E20,q−1

0

/ E∞ q,0

/ Eq q,0

dq

/ Eq 0,q−1

/ E∞ 0,q

/ 0

E2q,0 ∗,∗

Proposition B.4 Let fr : (Er , dr ) → (E ∗,∗ r , dr ) be a morphism between two spectral sequences such that fn is an isomorphism for some integer n. Then, for any integer m, n ≤ m, fm is an isomorphism. If we have first quadrant spectral sequences, then f also induces an isomorphism between E∞ and E ∞ .

B.2

Spectral sequences in cohomology

Definition B.5 A spectral sequence in cohomology Er , dr is a sequence, p,q

p+r,q−r+1

(Er ) of Z-bigraded differential modules such that dr : Er → Er and Er+1 = H(Er , dr ). A morphism of spectral sequences f : Er , dr → E r , dr is a family of morphisms fr : Er → E r of bigradation (0, 0) and such that fr+1 = H(fr ). In fact, a spectral sequence in cohomology can be viewed as a third quadrant (homology) spectral sequence and all that we said above is easily adapted to this situation. For instance, we say that a spectral sequence in

411

412

B : Spectral sequences p,q

cohomology is a first quadrant spectral sequence if Er q < 0. In this case the representation goes like this: O > > > > > · · · · >FNHFHNFHNN · > >FHFHNNN d2 > H > FFHHNNN > > FFHFHHNNN> > FFHH N& > · > FFHHH · > · · · > FFFHH > H > FFHH > > FFHHH> FF H> d4> F$ · ·> > FFF·> > · · FF > FF> FF> > FF> > #

= 0 for p < 0 or

· · · /

Proposition B.4 is still valid in this context.

B.3

Spectral sequences and filtrations

Most of examples of spectral sequences arise from a filtration on a complex. Definition B.6 A filtered differential R-module is a graded R-module M n n n+1 , F p+1 M ⊂ F p (M) and d(F p (M)) ⊂ such that M = ⊕∞ n=0 M , d(M ) ⊂ M p F (M). The filtration is called bounded if, for any n, there exist two integers s(n) and t(n) such that 0 = F s(n) (Mn ) ⊂ F s(n)−1 (Mn ) ⊂ · · · ⊂ F t(n) (Mn ) = Mn . Definition B.7 A bigraded differential algebra is a bigraded Z-module E∗,∗ with a multiplication Ep,q ⊗ Er,s → Ep+r,q+s and a differential d such that ⎛ ⎞ / / Ep,q ⎠ ⊂ Er,s . d⎝ p+q=n

r+s=n+1

A filtered differential graded algebra is a filtered differential module and a differential graded algebra A such that the multiplication sends F r A ⊗ F s A into F r+s A. Theorem B.8 Any filtered differential graded R-module gives rise to p,q ∗,∗ = a spectral sequence in cohomology Er , dr , r ≥ 1, with E1 p+q p p+1 H (F M/F M). If the filtration is bounded, the spectral sequence converges to H(M, d).

B.4

Serre spectral sequence

In the case of a filtered differential graded algebra, any stage of the spectral sequence is a bigraded differential algebra. Of course, this is where the use of upper and lower indices shows itself to be powerful. We set: • Zrp,q = F p Mp+q ∩ d −1 (F p+r Mp+q+1 ); p p+q ∩ d(F p−r Mp+q−1 ); • Bp,q r =F M p,q • Er = Zrp,q /(Zp+1,q−1 + dZp−r+1,q+r−2 ). r−1

r−1

With this notation, we have to prove that we get a spectral sequence. For p,q p+r,q−r+1 p,q instance, the differential d sends Zr to Zr and induces dr : Er → p+r,q−r+1 Er . We refer the reader to [186, Chapter 2].

B.4


Theorem B.9 ([236]) Let f : E → B be a fibration with F path connected and B simply connected. We consider homology and cohomology with coefficients in any commutative ring. 1. There is a filtration of Hn (E), 0 = H−1,n+1 ⊂ H0,n ⊂ H1,n−1 ⊂ · · · ⊂ Hn−1,1 ⊂ Hn,0 = Hn (E), and a first quadrant spectral sequence such that: ∼ E2p,q = Hp (B; Hq (F)), E∞ p,q = Hp,q /Hp−1,q+1 . 2. There is a filtration on H n (E) 0 = H n+1,−1 ⊂ H n,0 ⊂ · · · ⊂ H 1,n−1 ⊂ H 0,n = H n (E), and a first quadrant spectral sequence in cohomology such that: p,q

p,q

E2 = H p (B; H q (F)), E∞ = H p,q /H p+1,q−1 . Each (Er , dr ) is a differential bigraded algebra. This spectral sequence is called the Serre spectral sequence. The key to its construction is a filtration of the chain complex on the total space by a “basic degree”. For more general versions using local coefficients, see [236]. Example B.10 In general, a spectral sequence gives little information about the algebra structure of H(A, d) because the convergence uses an associated graded step. We give here an illustration of this fact, see also Remark 3.36. Consider the Hopf fibration S3 → S7 → S4 . By taking the orbits of the S1 action on the fiber and the total space, we get a fibration S2 → CP(3) → S4 .

413

414


The Serre spectral sequence of this fibration has E2 = H p (S4 ) ⊗ H q (S2 ) and thus collapses at level 2. Yet, the cohomology algebra of CP(3) is not the tensor product of these two algebras. In this example, we can also observe that the Euler class is zero and that the fibration has no section. In opposition with the previous example, the next one shows how we can get information about the cohomology algebra of the loop space S3 . Example B.11 Consider the fibration S3 → PS3 → S3 . The cohomology ring H ∗ (S3 ; Z) has a single generator x2n in each even degree with the only relations being n · x2n = x2 n . Thus, all powers of x2 are nontrivial. Proof The fibration S3 → PS3 → S3 , where PS3 = {γ : I → S3 | γ (0) = (1, 0, 0, 0)} is the contractible path space on S3 , has the following spectral sequence diagram. x6 K x6 ⊗ y3 KK KK KK KK KK KK KK KK KK KK KK K% x4 K x4 ⊗ y3 KK KK KK KK KK KK KK KK KK KK KK K% x2 L x2 ⊗ y3 LL LL LL LL LL LL LL LL LL LL LL LL % y3 Here we are not saying that we know the cohomology of S3 . We are simply indicating the structure that we now determine algebraically for the convenience of the reader. The diagram has two vertical lines with the cohomology of the fiber along the vertical axis and the cohomology of the base along the horizontal axis. The only possible nonzero differential is the one shown, d3 , which moves over to the right by three and moves down by two. Because the path

B.4


space is contractible, the class y3 cannot survive to infinity, for it would then provide a nonzero class in H 3 (PS3 ; Z) = 0. Thus, there must exist an element x2 with d3 (x2 ) = y3 . The differential d3 is a derivation, so we have d3 (x2 2 ) = 2x2 ⊗ d3 (x2 ) = 2x2 ⊗ y3 . This says that the element x2 ⊗ y3 must still be killed in order to avoid producing a nonzero class in H 5 (PS3 ; Z) = 0. So there must exist x4 with d3 (x4 ) = x2 ⊗ y3 . Hence d3 (2x4 − x2 2 ) = 0. The cocycle 2x4 − x2 2 would survive to infinity if it were not zero, so we must have 2x4 = x2 2 . We will do one more case to see the general algebraic pattern. We have d3 (x2 3 ) = 3x2 2 ⊗ y3 = 6x4 ⊗ y3 , so there must exist x6 with d3 (x6 ) = x4 ⊗ y3 . Therefore, d3 (3 · x6 ) = d3 (x2 3 ) and, by the same argument as before, we must have 3 · x6 = x2 3 . Inductively, we can then show that classes x2n exist with n · x2n = x2 n . We now state three particular cases of the Serre spectral sequence. Proofs can be found in any classical algebraic topology book or in [186]. Corollary B.12 With the hypotheses of Theorem B.9, suppose that Hp (B) = 0 for 0 < p < p0 and Hq (F) = 0 for 0 < q < q0 . Then there is an exact sequence Hp0 +q0 −1 (F) →Hp0 +q0 −1 (E) →Hp0 +q0 −1 (B) →Hp0 +q0 −2 (F) · · · →H1 (E) → 0. Corollary B.13 (Gysin sequence) With the hypotheses of Theorem B.9, suppose that the fiber F has the homology of a sphere Sm . Then we have exact sequences, ···

/ Hk+1 (E)

/ Hk+1 (B)

/ Hk−m (B)

/ Hk (E)

/ ··· ,

/ H k+1 (E)

/ ···

and ···

/ H k (E)

/ H k−m (B)

ψ

/ H k+1 (B)

The morphism ψ satisfies ψ(v1 ∪v2 ) = v1 ∪ψ(v2 ). In fact, ψ is multiplication by the Euler class of the fibration. Corollary B.14 (Wang sequence) With the hypotheses of Theorem B.9, suppose that the basis B has the homology of a sphere Sm . Then we have two exact sequences, ···

/ Hk (F)

···

/ H k−1 (F)

and

/ Hk (E) θ

/ H k−m (F)

/ Hk−m (F) / H k (E)

The morphism θ acts as an algebra derivation.

/ Hk−1 (F)

/ ··· ,

/ H k (F)

/ ··· .

415

416

B : Spectral sequences

B.5

Zeeman–Moore theorem

Since we are concerned with rational homotopy type, we now work with vector spaces over a field of characteristic zero. ∗,∗

Theorem B.15 (Zeeman–Moore theorem) If fr : (Er , dr ) → (E ∗,∗ r , dr ) is a p,q morphism between two first quadrant spectral sequences such that E2 = p,0

0,q

p,q

p,0

0,q

p,q

E2 ⊗ E2 , E 2 = E 2 ⊗ E 2 and f2 following conditions imply the third: p,0

p,0

0,q

0,q

p,0

= f2

0,q

⊗ f2 , then any two of the

(1) f2 : E2 → E 2 is an isomorphism for all p;

(2) f2 : E2 → E 2 is an isomorphism for all q; p,q (3) f∞ : E∞ → E p,q ∞ is an isomorphism for all (p, q). This theorem was first proved by J.C. Moore [200, Exposé 22]. Generalizations were given by Zeeman [268], see also [130, Theorem 17.17]. From Theorem B.9 we immediately obtain the following topological interpretation. Corollary B.16 Let f

F

/ F

E

g

/ E

B

h

/ B

be a morphism between orientable fibrations. Then any two of the following conditions imply the third: (1) H ∗ (h) is an isomorphism; (2) H ∗ (f ) is an isomorphism; (3) H ∗ (g) is an isomorphism. Proof In the statement of Theorem B.15, if properties (1) and (2) are satisfied, then (3) will follow directly from the hypotheses and Proposition B.4. We suppose now that (1) and (3) are satisfied and prove (2). We follow the proof of [182, Theorem 11.1, Chapter XI] with an appropriate modification in the case of cohomology. We will use the classical Four Lemma of homological algebra that we recall below for the convenience of the reader, see Lemma B.17. We consider the property: 0,q 0,q (2)j : the map f2 : E2 → E 2 is an isomorphism for all q, 0 ≤ q ≤ j.

B.5 0,0

Zeeman–Moore theorem

0,0

Observe that E2 = E∞ . The hypothesis (3) implies that (2)0 is satisfied. So we now work by induction and assume (1), (3) and (2)j , with j fixed. Step 1. We do a second induction, on the integer i, considering the property j

p,q

(2)i : fi

is an isomorphism for all q, q + i ≤ j + 2, and all p, and a monomorphism for all q, q ≤ j, and all p. j

j

Observe that (2)2 is true and suppose that (2)n is satisfied for some integer j

n. We have to prove (2)n+1 and for that we need to come back to the construction of the (n + 1)st page from the nth page. First we look at the ∗,∗ subspaces Zn . Consider the following morphism between exact sequences: / Zp,q n

0

/ Ep,q n

p,q

p,q

Z(f )n

fn

/ Z p,q n

0

/ Ep+n,q−n+1 n

/ E p,q n

p+n,q−n+1

fn

/ E p+n,q−n+1 n p+n,q−n+1

If q ≤ j, then (q − n + 1) + n ≤ q + 1 ≤ j + 2 and fn is an p,q p,q isomorphism and fn a monomorphism, by assumption. Therefore, Z(f )n is a monomorphism. p,q p+n,q−n+1 If q ≤ j − n + 1, then fn is an isomorphism and fn a p,q monomorphism. Therefore Z(f )n is an isomorphism. ∗,∗ We look now at the Bn : / Bp,q n

p−n,q+n−1

En

p,q

p−n,q+n−1

fn

/ 0

B(f )n

/ B p,q n

E p−n,q+n−1 n

/ 0

p−n,q+n−1

p,q

is an isomorphism. Therefore B(f )n If q ≤ j, then fn surjective. ∗,∗ Finally, we arrive at the space En+1 : 0

/ Bp,q n p,q

B(f )n

0

/ B p,q n

/ Zp,q n

/ Ep,q n+1 p,q

p,q

Z(f )n

/ Z p,q n

/ 0

fn+1

/ E p,q n+1

/ 0

is

417

418


p,q

If q ≤ j, Z(f )n is a monomorphism and B(f )n is surjective. Therefore is a monomorphism. p,q If q ≤ j − (n + 1) + 2 = j − n + 1, then Z(f )n is an isomorphism. p,q Therefore, fn+1 is an isomorphism.

p,q fn+1

j

We have thus proved (2)i for any i. Step 2. We establish (2)j+1 . In order to see this, using a descending induction 0,j+1 0,j+1 0,j+1 on i, we prove that fi : Ei → E i is an isomorphism for any i ≥ 2. Observe that, in first quadrant spectral sequences, we have: 0,j+1 ∼ 0,j+1 – Ei = E∞ for i large enough, 0,j+1

−i,j+i

0,j+1

0,j+1

= Ei+1 , for any i ≥ 2, because the range of di : Ei → Ei – Zi is zero. 0,j+1 From this first property and the hypothesis (3), we may assume that fn+1 0,j+1

is an isomorphism and we are reduced to proving that fn is an isomorphism also. We consider the following morphism between two exact sequences: / Z0,j+1 n

0

0,j+1

Z(f )n

/ En,j−n+2 n

0,j+1

fn

/ Z 0,j+1 n

0

dn

/ E0,j+1 n

n,j−n+2

/ E 0,j+1 n

dn

fn

/ E n,j−n+2 n

n,j−n+2

From Step 1, we know that fn is an isomorphism. On the other hand, 0,j+1 0,j+1 0,j+1 we have Zn = En+1 , so Z(f )n is an isomorphism by induction. This 0,j+1

implies the injectivity of fn . n,j−n+2 is surjective; this implies the surjectivity of Recall now that B(f )n 0,j+1 n,j−n+2 n,j−n+2 fn by the same argument, replacing En by Bn in the previous diagram. 0,j+1 We then get that fn is an isomorphism and (2)j+1 follows. The same argument works to prove that (2) and (3) imply (1). Lemma B.17 (The Four Lemma [182, Lemma 3.3, Chapter I]) Consider the following commutative diagram with exact rows. / •

•

•

α1

/ •

/ • α2

/ •

/ • α3

/ •

/ • α4

α5

/ •

(i) If α1 is an epimorphism and α2 and α4 monomorphisms, then α3 is a monomorphism.

B.6

An algebraic example: The odd spectral sequence

(ii) If α5 is a monomorphism and α2 and α4 epimorphisms, then α3 is an epimorphism.

B.6

An algebraic example: The odd spectral sequence

The following result gives the structure of the cohomology algebra H ∗ (G/T; Q) where T is the maximal torus of a compact connected Lie group G and, more generally, for the algebra H ∗ (G/H; Q) if H is a closed subgroup of maximal rank in G, see Theorem 3.33. Theorem B.18 Let (A = ∧(Q ⊕ P), d) be a cdga where Q is concentrated in even degrees and P in odd degrees such that d(Q) = 0 and d(P) ⊂ ∧Q. Further, suppose that H(A, d) is of finite dimension. Then the following conditions are equivalent: (1) P and Q have the same dimension; (2) χ (H(A, d)) > 0; (3) H(A, d) is evenly graded. If these conditions are satisfied, then H(A, d) = ∧Q/I, where I is the ideal generated by dP. This statement can be generalized to any minimal model, see [129] or [87, page 444]. Proof We give A a new (lower) gradation defined by Gs A = ⊕i (∧Q ⊗ ∧i P)s−i and observe from the hypotheses that dGs A ⊂ Gs A. Therefore, this gradation induces a gradation Gs H on cohomology H = H(A, d) and the corresponding Euler characteristics of Gs H and Gs A are equal. Observe also that an element of (∧Q ⊗ ∧i P) is of odd lower degree if and only if i is odd. As a consequence, we have: (−1)i dim H((∧Q ⊗ ∧i P)s−i , d) χ(Gs H) = i

(−1)i dim (∧Q ⊗ ∧i P)s−i . = i

Therefore, the global Euler characteristic of H = ⊕s Gs H is obtained by χ (H) = χ(Gs H) = (−1)i dim (∧Q ⊗ ∧i P)s−i . s

s

i

To use the finite dimensional hypothesis, we have to take into account a special type of Poincaré polynomial adapted to this situation called the

419

420


Koszul–Poincaré polynomial:

P H(t) =

∞ s=0

χ(Gs H)t s =

∞ (−1)i dim (∧Q ⊗ ∧i P)s−i t s . s=0

i

This last term does not depend on the differential and can be computed as: 5p (1 − t 2ui ) P H(t) = 5i=1 , q 2vj j=1 (1 − t ) if Q is of dimension q generated by elements of degrees (2v1 , . . . , 2vq ) and P of dimension p generated by elements of degrees (2u1 −1, . . . , 2up −1). In order to have H of finite dimension, this expression must be a polynomial and we therefore require p ≥ q. 5 u If χ (H) > 0 then one has p = q and, in this case, χ (H) is the quotient 5 vi . j

Reciprocally, p = q implies χ(H) > 0. We have thus proved the equivalence between (1) and (2). Obviously (3) implies (2). So we are reduced to proving (1) or (2) implies (3). For that, we use a classical construction in minimal models in which we kill the vector space Q by adding a copy sQ, shifted by one degree (i.e. |sq| = |q| − 1 if q ∈ Q). We construct a cdga (∧Q ⊗ ∧P ⊗ ∧sQ, d) with d(sq) = q, observe that (∧Q ⊗ ∧sQ, d) is acyclic and therefore see that the cohomology of (∧Q ⊗ ∧P ⊗ ∧sQ, d) is ∧P. We now forget all the degrees and consider the degree in word length in P and sQ. In (∧Q⊗∧P ⊗∧sQ, d), the differential d lowers this word length by exactly one, so we have an induced lower degree on the cohomology. We denote it by Hk . We now filter (∧Q ⊗ ∧P ⊗ ∧sQ, d) by the word length degree in sQ. This i,j gives a spectral sequence with first page E1 = ⊕i (∧i sQ) ⊗ Hj (∧Q ⊗ ∧P). Let k be the greatest integer such that Hk (∧Q ⊗ ∧P) = 0 and let α ∈ Hk . Choose an element a of maximal word length in ∧sQ that is of degree dim Q. (For this, recall that Q is in even degrees, so sQ is in odd degrees. Therefore, the product of all generators in sQ is one such element a.) The product a α cannot be killed in the spectral sequence (since it is in the corner), so dim P ≥ k + dim Q. From the assumption dim P = dim Q, we deduce k = 0 and, thus, H = ∧Q/(dP) is evenly graded.

B.7

A particular case: A double complex

Definition B.19 A double complex is a bigraded R-module M equipped with two differentials, d : Mp,q → Mp+1,q and d : Mp,q → Mp,q+1 , such that

B.7

A particular case: A double complex

d d + d d = d 2 = d 2 = 0. The total complex associated to the double complex is (Tot(M), d) with Tot(M)n = ⊕p+q=n Mp,q and d = d + d . Since d 2 = d 2 = 0, we have two cohomologies associated to a double complex: ∗,∗

∗,∗

HI (M) = H(M∗,∗ , d ) and HII (M) = H(M∗,∗ , d ). Because d d + d d = 0, the respective differentials d and d induce differentials d¯ on HII (M∗,∗ ) and d¯ on HI (M∗,∗ ) so that we get two new cohomologies HI HII (M) = H(HII (M), d¯ ) and HII HI (M) = H(HI (M), d¯ ). ∗,∗

∗,∗

∗,∗

∗,∗

Theorem B.20 Let (M∗,∗ , d , d ) be a doublecomplex. Then we have two spectral sequences: I Er , I dr and II Er , II dr such that ∗,∗ I E2

∗,∗

∗,∗

∗,∗

= HI HII (M) and II E2 = HII HI (M).

Moreover, if Mp,q = 0 when p < 0 or q < 0, then the two spectral sequences converge to H ∗ (Tot(M), d). The idea is to take the two spectral sequences associated to the following filtrations: / / p p FI (Tot(M))t = Mr,t−r and FII (Tot(M))t = Mt−r,r . r≥p

r≥p

A quick (and wrong!) glimpse at this situation might elicit the idea that such spectral sequences always collapse at level 2. A counterexample is given in the case of the Frölicher spectral sequence in Example 4.75. Exercises for Appendix B In order to understand spectral sequences, the reader needs to experiment and do concrete examples on his or her own. Here we suggest some “easy” problems for a beginner. Exercise B.1 Use the Wang exact sequence to prove Hq (Sm ) = Z if q = k(m − 1) and 0 otherwise. Exercise B.2 Give a generalization of Example B.11 and determine the cohomology algebra of Sm . Hint: see for instance [37, page 204]. Exercise B.3 Let f : S3 → K(Z, 3) be the classifying map of the fundamental class. Let X denote the pullback of the path fibration PK(Z, 3) → K(Z, 3) along f . Compute H 5 (X). Exercise B.4 By using the Serre spectral sequence of the fibration S1 → S2n+1 → CP(n), determine the cohomology algebra of H ∗ (CP(n); Z). (Remember, the differentials are derivations.)

421

422


Exercise B.5 By using the Serre spectral sequence of the path fibration K(Z, n) → PK(Z, n + 1) → K(Z, n + 1), determine the cohomology algebra H ∗ (K(Z, n); Q). Exercise B.6 By using the Serre spectral sequence of the fibration U(n) → EU(n) → BU(n), determine the cohomology algebra H ∗ (BU(n); Q). Exercise B.7 Compute the cohomology algebras, with coefficients in Q, of the various Stiefel and Grassmann manifolds.

Appendix C Basic homotopy recollections

In this appendix we collect a few classical notions and results in homotopy theory that are used in various chapters. We denote the set of homotopy classes of maps from the (pointed) space X to the (pointed) space Y by [X, Y], where we take homotopies relative to the basepoint. If f : X → Y, then pre- and post-composition with f induce respective morphisms of homotopy sets f : [Y, Z] → [X, Z] and f : [Z, X] → [Z, Y].

C.1 n-equivalences and homotopy sets In Chapter 2, we defined a notion of cdga homotopy and considered sets of homotopy classes of cdga morphisms. We saw there that rational homotopy equivalences (i.e. quasi-isomorphisms) induce bijections of homotopy sets. This result is the cdga analogue of the following thread of ideas. Definition C.1 ([240, page 404]) A continuous map of path connected spaces f : X → Y is called an n-equivalence for n ≥ 1 if for every x ∈ X, the induced map f = πq (f ) : πq (X, x) → πq (Y, f (x)) is an isomorphism for 0 < q < n and an epimorphism for q = n. A weak equivalence or ∞-equivalence is an n-equivalence for all n ≥ 1. The effect of an n-equivalence on homotopy sets is described by the Proposition C.2 ([240, Corollary 7.6.23]) If f : X → Y is an n-equivalence, then – f : [P, X] → [P, Y] is bijective if P is a CW-complex of dimension ≤ n − 1; – f : [P, X] → [P, Y] is surjective if P is a CW-complex of dimension ≤ n. The next result, Whitehead’s theorem, provides a way to detect nequivalences of simply connected spaces by looking at homology.

424

C : Basic homotopy recollections

Theorem C.3 ([240, Theorem 7.5.9]) Let f : (X, x0 ) → (Y, y0 ) be a map of path connected and pointed spaces. If there is n ≥ 1 such that πq (f ) : πq (X, x0 ) → πq (Y, y0 ) is an isomorphism for q < n and an epimorphism for q = n, then Hq (f ) : Hq (X, x0 ) → Hq (Y, y0 ) is an isomorphism for q < n and an epimorphism for q = n. Conversely, if X and Y are simply connected and Hq (f ) is an isomorphism for q < n and an epimorphism for q = n then f is an n-equivalence. A map f : X → Y is a homotopy equivalence if there exists g : Y → X with fg idY and gf idX . The symbol denotes the homotopy relation between maps and we also write X Y when there exists a homotopy equivalence X → Y. We then say that X and Y have the same homotopy type. Note that we do not need a chain of maps here as in the definition of rational homotopy type (see Definition 2.34) because in this case we require the existence of a map that is a homotopy inverse. The following result provides the link between weak equivalences and homotopy equivalences. Theorem C.4 ([240, Corollary 7.6.24]) A map f : X → Y of CW complexes is a weak equivalence if and only if it is a homotopy equivalence.

C.2

Homotopy pushouts and pullbacks

Throughout the book, we have used the notions of homotopy pushout and homotopy pullback. In this section, we recall these general notions and mention several simple results about them. Given two maps f : X → W and g : Y → W, form the space P(f , g) = {(x, θ, y) ∈ X × W I × Y) | f (x) = θ (0), g(y) = θ (1)}. This space is the homotopy pullback of f and g in the following sense. First, the square

P(f , g) f¯

g¯

Y

/ X

g

/ W

f

C.2


is homotopy commutative, where f¯ and g¯ are the canonical projections. Further, given any other homotopy commutative square

h

Z k

/ X

Y

g

f

/ W

there exists a map ω : Z → P(f , g)

Z

h

ω

k

" P(f , g)

g¯

%

/ X

f¯

Y

g

f

/ W

in which gω ¯ h : Z → X and f¯ ω k : Z → Y. The map ω is not unique, even up to homotopy. The following are fundamental results for homotopy pullbacks.

Proposition C.5 ([184]) In the following diagram,

A

/ B

/ C

D

/ E

/ F

the squares BCFE and ACFD are homotopy pullbacks if and only if the squares BCFE and ABED are homotopy pullbacks.

425

426


Lemma C.6 ([184, Corollary 7, Corollary 9]) Consider the homotopy commutative cube below: U1

/ X1

BB BB BB BB

h

Y1

k

g

U2

/ X2

BB BB BB BB

X2

BB BB BB BB / V1

f

BB BB BB BB / V2

Suppose that the top and bottom squares are homotopy pullbacks. If the maps f , g and h are homotopy equivalences, then so is the map k. The dual notion is that of a homotopy pushout. Given maps f : W → X and g : W → Y, the homotopy pushout of f and g is given by D(f , g) = (X ∪ (W × I) ∪ Y)/ ∼ where (w, 0) ∼ f (w) and (w, 1) ∼ g(w). This space is also called the the double mapping cylinder of f and g and is a homotopy pushout of f and g in the following sense. First, the square f

/ X

W g

Y

g¯

/ D(f , g)

f¯

is homotopy commutative, where f¯ and g¯ are the canonical inclusions. Further, given any other homotopy commutative square f

W g

Y

/ X

k

/ Z

h

C.2


there exists a map ω : D(f , g) → Z f

/ X

W

g¯

g

Y

f¯

/ D(f , g)

h ω

" ,

k

Z

in which ωg¯ h : X → Z and ωf¯ k : Y → Z. In general, the map ω is not unique, even up to homotopy. Two basic results about homotopy pushouts are the following. Proposition C.7 In the following diagram, A

/ B

/ C

D

/ E

/ F

the squares ABED and ACFD are homotopy pushouts if and only if the squares ABED and BCFE are homotopy pushouts. Lemma C.8 ([184, Corollary 7, Corollary 9]) Consider the following homotopy commutative cube: U1

/ X1

BB BB BB BB

h

Y1

f

U2

g

BB BB BB BB

X2

/ X2

BB BB BB BB / V1

k

BB BB BB BB / V2

427

428


Suppose that the top and bottom squares are homotopy pushouts. If the maps f , g and h are homotopy equivalences, then so is the map k. As indicated by the two statements above, properties of homotopy pullbacks can often be dualized for homotopy pushouts. Although the next result is a case where duality between homotopy pullbacks and pushouts fails, it nevertheless provides a useful homotopical tool. Theorem C.9 (Theorem of the cube) If the bottom of a cube is a homotopy pushout and all sides are homotopy pullbacks, then the top of the cube is a homotopy pushout. As a consequence, we have Corollary C.10 Let X be a space and ABED be a homotopy pushout. Then the square A×X

/ B×X

/ E×X

D×X is a homotopy pushout.

We note here that homotopy pullbacks and homotopy pushouts are really only defined up to homotopy equivalence. Replacing any of the spaces or maps by homotopy equivalent ones in the initial diagram produces the same homotopy pushout or pullback up to homotopy. In some discussions of homotopy pushouts and pullbacks, such as in [184], the specific homotopies being used are also taken into account.

C.3

Cofibrations and fibrations

Two of the most basic homotopical objects are fibrations and cofibrations. Fibrations are probably more familiar to geometers because they are the homotopical analogues of fiber bundles. Indeed, locally trivial bundles with paracompact base spaces and covering spaces are examples of fibrations. We begin by recalling the notion of cofibration and then see how duality provides the definition of fibration. For this section, we refer to [137] and [265].

C.3


Definition C.11 A map j : A → X is a cofibration if for any commutative diagram of solid arrows

A j

X

id×0

/ A×I

j×id

id×0

/ X×I

H H

f

" ,

Z

there exists a dotted arrow H such that the diagram still commutes. The quotient map q : X → X/j(A) (or the space X/j(A)) is called the cofiber of j. If j : A → X is a cofibration, the map j is injective and induces a homeomorphism from A to j(A). Therefore, we consider A as a subset of X (closed in X if X is Hausdorff) and denote the cofiber by X/A. For a space A, let CA denote the cone on A defined by CA = (A × I)/((a, 1) ∼ ∗). If f : A → X is any map, recall that the mapping cone of f (or the homotopy cofiber of f ) is the inclusion q : X → Cf = X ∪f CA (or the space Cf = X ∪f CA) with X ∪f CA = X ∪ CA/((f (a) ∼ (a, 0)). In the case of a cofibration j : A → X, the cofiber X/A has the homotopy type of the cone Cj . That is, we have a homotopy equivalence of pairs (X/A, ∗) (Cj , CA). A space X has a non-degenerate basepoint x0 if the inclusion x0 → X is a cofibration. This implies, at least for compactly generated spaces (see [265] for this definition), that there is an open neighborhood U of x0 which contracts to x0 in X relative to x0 . In other words, there is a homotopy H : U × I → X such that H(u, 0) = u, H(u, 1) = x0 and H(x0 , t) = x0 . CW complexes always have nondegenerate basepoints. Indeed, every point of a CW complex is a nondegenerate basepoint. A space is said to be well-pointed if it has a nondegenerate basepoint. If the spaces are well-pointed, A → A ∨ X is a cofibration, called a trivial cofibration, with cofiber X. Note also that the inclusion A → CA is a cofibration with cofiber A. More generally, if f : A → X is any map, then the mapping cylinder A → Mf = X ∪ (A ∧ I)/(f (a) ∼ a ∧ 0), a → a ∧ 1, is a cofibration with cofiber Cf and f is the composition of a cofibration and

429

430


a homotopy equivalence: f

/ X > ~ ~ ~ ~~ ~~

A @ @@ @@ @@ jf Mf f

q

Any sequence A −→ X −→ C is called a cofiber sequence if q : X → C has the homotopy type of Mf −→ X −→ Cf . For any map f : A → X, we have the Barratt–Puppe sequence, f

f

δ

q

q

A −→ X −→ Cf −→ A −→ X −→ Cf −→ · · · where each three-term sequence is a cofiber sequence. For any space Z, the Barratt–Puppe sequence induces an exact sequence of pointed sets or groups: f

q

δ

[A, Z] ←− [X, Z] ←− [Cf , Z] ←− [ A, Z] ( f )

( q)

←− [ X, Z] ←− [ Cf , Z] ←− · · ·

Finally, we mention that, if either of the maps f or g which define a homotopy pushout is a cofibration, then the homotopy pushout of f and g has the homotopy type of the ordinary topological pushout of f and g. Now suppose that, instead of considering the problem of extending homotopies, we dualize the definition of cofibration and consider the problem of lifting homotopies. With this in mind, for a space Y, let Y I = {ω : [0, 1] → Y} be the free path space on Y. Note that any map f : X → Y induces f I : X I → Y I , ω → f ◦ ω. Definition C.12 A map p : E → B is a fibration if for any commutative diagram of solid arrows Z

G

G

g

pI

EI

E

ev0

# / BI

p

/ B

ev0

C.3


there exists a dotted arrow G such that the diagram still commutes. Here, for any space Y, the notation ev0 denotes an evaluation mapping Y I → Y given by ev0 (ω) = ω(0). The map F = p−1 (∗) → E (or the space F) is called the fiber of p. Note that the diagram above is equivalent to the following one defining the homotopy lifting property. The G’s in the following diagram have tildes over them to denote that they are the adjoints of the maps in Definition C.12 under the exponential correspondence. Z×0

g

G

Z×I

/ E <

G

p

/ B

Namely, given the commutative diagram of solid arrows, it is the homotopy . lifting property which ensures the existence of the lift G If p : E → B is a fibration with B path connected, the map p is onto. If f : X → Y is any map, the homotopy fiber of f is the map Ff → X (or the space Ff ) defined by Ff = (ω, x) ∈ Y I × X | f (x) = ω(0) and ω(1) = ∗ , (ω, x) → x. In the case of a fibration p, the fiber F has the homotopy type of the homotopy fiber Fp . The projection pB : E × B → B is a fibration, called the trivial fibration, with fiber E. The path fibration P(X) = ω ∈ X I | ω(0) = ∗ → X, ω → ω(1), is a fibration with the loop space X as fiber. More generally, if f : E → B is any map, the path fibration associated to f , pf : Ef → B, is a fibration with fiber Ff , defined by Ef = (ω, x) ∈ BI × E | f (x) = ω(0) , pf ((ω, x)) = ω(1). Any map is the composition of a homotopy equivalence and a fibration as shown below. f

E ? ?? ?? ??

/ B ? pf

Ef Note that, as we mentioned at the start of Section C.3, locally trivial bundles with paracompact base spaces and covering spaces are fibrations. ι

f

Any sequence F −→ E −→ B is called a fiber sequence if ι : F → E has the homotopy type of Ff → Ef → E. For any map f : E → B we have the

431

432


Puppe sequence f

ι

ι

∂

f

. . . −→ Ff −→ E −→ B −→ Ff −→ E −→ B where each three-term sequence is a fiber sequence. For any space Z, the Puppe sequence induces an exact sequence of pointed sets or groups: f

ι

∂

(ι)

[Z, B] ←− [Z, E] ←− [Z, Ff ] ←− [Z, B] ←− [Z, E] ←− · · · Recall that the multiplication in the space of loops B, µ : B × B → B, is the ordinary composition of loops and the inverse map ν : B → B is given by ν(ω)(t) = ω(1 − t), ω ∈ B. The group-like properties of B extend to the analogue of a group action of B on the fiber F. This is expressed in the following Definition C.13 Let p : E→B be a fibration of fiber F. The holonomy of p is a map Hol : B × F → F obtained by the homotopy lifting property applied to G(ω, x, t) = ω(t): / E :

B × F × 0 G

B × F × I

G

p

/ B

with Hol(ω, x) = G (ω, x, 1). Finally, we mention that, if either of the two maps f or g defining a homotopy pullback is a fibration, then the homotopy pullback of f and g is of the homotopy type of the topological pullback of f and g. Of course, the topological pullback is both easier to define and more familiar from its role in bundle theory, so this identification is often useful.

References

[1] Adams, J. Frank (1957). On the cobar construction. In Colloque de topologie algébrique, Louvain, 1956, pp. 81–87. Georges Thone, Liège. [2] Adams, J. Frank (1969). Lectures on Lie Groups. W. A. Benjamin, Inc., New York. [3] Allday, Christopher (1978). On the rational homotopy of fixed point sets of torus actions. Topology, 17(1), 95–100. [4] Allday, Christopher (1994). Circle actions on Kähler spaces. J. Pure Appl. Algebra, 91(1-3), 23–27. [5] Allday, Christopher (1998). Examples of circle actions on symplectic spaces. In Homotopy and Geometry (Warsaw, 1997), Volume 45 of Banach Center Publ., pp. 87–90. Polish Acad. Sci., Warsaw. [6] Allday, Christopher and Halperin, Stephen (1978). Lie group actions on spaces of finite rank. Quart. J. Math. Oxford Ser. (2), 29(113), 63–76. [7] Allday, Christopher and Puppe, Volker (1985). On the Localization Theorem at the Cochain Level and Free Torus Actions, Volume 1172 of Lecture Notes in Math., pp. 1–16. Springer, Berlin. [8] Allday, Christopher and Puppe, Volker (1986). Bounds on the Torus Rank, Volume 1217 of Lecture Notes in Math., pp. 1–10. Springer, Berlin. [9] Allday, Christopher and Puppe, Volker (1986). On the rational homotopy Lie algebra of a fixed point set of a torus action. Trans. Amer. Math. Soc., 297(2), 521–528. [10] Allday, Christopher and Puppe, Volker (1993). Cohomological Methods in Transformation Groups, Volume 32 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge. [11] Allday, Christopher and Puppe, Volker (2006). The minimal Hirsch– Brown model via classical Hodge theory. Pacific J. Math., 226(1), 41–51. [12] Anosov, Dmitri Victorovich (1982). Generic properties of closed geodesics. Izv. Akad. Nauk SSSR Ser. Mat., 46(4), 675–709, 896.

434

References

[13] Aouina, Mokhtar and Klein, John R. (2004). On the homotopy invariance of configuration spaces. Algebr. Geom. Topol., 4, 813–827. [14] Atiyah, Michael (1957). Complex analytic connections in fiber bundles. Trans. Amer. Math. Soc., 85, 181–207. [15] Atiyah, Michael and Bott, Raoul (1984). The moment map and equivariant cohomology. Topology, 23(1), 1–28. [16] Audin, Michèle (1991). Exemples de variétés presque complexes. Enseign. Math. (2), 37(1-2), 175–190. [17] Babenko, Ivan K. and Taimanov, Iskander A. (2000). On nonformal simply connected symplectic manifolds. Sibirsk. Mat. Zh., 41(2), 253–269, i. [18] Ballmann, Werner and Ziller, Wolfgang (1982). On the number of closed geodesics on a compact Riemannian manifold. Duke Math. J., 49(3), 629–632. [19] Bangert, Victor (1993). On the existence of closed geodesics on twospheres. Internat. J. Math., 4(1), 1–10. [20] Bangert, Victor and Hingston, Nancy (1984). Closed geodesics on manifolds with infinite abelian fundamental group. J. Differential Geom., 19(2), 277–282. [21] Barge, Jean (1976). Structures différentiables sur les types d’homotopie rationnelle simplement connexes. Ann. Sci. École Norm. Sup. (4), 9(4), 469–501. [22] Belegradek, Igor and Kapovitch, Vitali (2001). Topological obstructions to nonnegative curvature. Math. Ann., 320(1), 167–190. [23] Belegradek, Igor and Kapovitch, Vitali (2003). Obstructions to nonnegative curvature and rational homotopy theory. J. Amer. Math. Soc., 16(2), 259–284. [24] Bendersky, Martin and Gitler, Sam (1991). The cohomology of certain function spaces. Trans. Amer. Math. Soc., 326(1), 423–440. [25] Benson, Chal and Gordon, Carolyn S. (1988). Kähler and symplectic structures on nilmanifolds. Topology, 27(4), 513–518. [26] Berceanu, Barbu, Markl, Martin, and Papadima, Stefan. ¸ (2005). Multiplicative models for configuration spaces of algebraic varieties. Topology, 44(2), 415–440. [27] Besse, Arthur L. (1978). Manifolds all of whose Geodesics are Closed, Volume 93 of Ergebnisse der Mathematik und ihrer Grenzgebiete [Results in Mathematics and Related Areas]. Springer-Verlag, Berlin. With appendices by D. B. A. Epstein, J.-P. Bourguignon, L. Bérard-Bergery, M. Berger and J. L. Kazdan.

References

[28] Bishop, Richard L. and Crittenden, Richard J. (1964). Geometry of Manifolds. Pure and Applied Mathematics, Vol. XV. Academic Press, New York. [29] Blanchard, André (1956). Sur les variétés analytiques complexes. Ann. Sci. Ecole Norm. Sup. (3), 73, 157–202. [30] Block, Jonathan and Lazarev, Andrej (2005). André–Quillen cohomology and rational homotopy of function spaces. Adv. Math., 193, 18–39. [31] Bödigheimer, Carl-Friedrich, Cohen, Frederick R., and Taylor, Laurence R. (1989). On the homology of configuration spaces. Topology, 28(1), 111–123. [32] Borel, Armand (1949-50). Exposés 12-13. Séminaire Cartan, Paris. [33] Borel, Armand (1960). Seminar on Transformation Groups. With contributions by G. Bredon, E. E. Floyd, D. Montgomery, R. Palais. Annals of Mathematics Studies, No. 46. Princeton University Press, Princeton, N.J. [34] Borel, Armand and Hirzebruch, Friedrich (1958). Characteristic classes and homogeneous spaces. I. Amer. J. Math., 80, 458–538. [35] Borel, Armand and Serre, Jean-Pierre (1953). Groupes de Lie et puissances réduites de Steenrod. Amer. J. Math., 75, 409–448. [36] Bott, Raoul (1956). An application of the Morse theory to the topology of Lie-groups. Bull. Soc. Math. France, 84, 251–281. [37] Bott, Raoul and Tu, Loring W. (1982). Differential Forms in Algebraic Topology, Volume 82 of Graduate Texts in Mathematics. Springer-Verlag, New York. [38] Bousfield, Aldridge K. and Gugenheim, V. K. A. M. (1976). On PL de Rham theory and rational homotopy type. Mem. Amer. Math. Soc., 8(179), ix+94. [39] Bousfield, Aldridge K., Peterson, Chariya, and Smith, Larry (1989). The rational homology of function spaces. Arch. Math. (Basel), 52(3), 275–283. [40] Bredon, Glen E. (1972). Introduction to Compact Transformation Groups. Academic Press, New York. Pure and Applied Mathematics, Vol. 46. [41] Bredon, Glen E. (1993). Topology and Geometry, Volume 139 of Graduate Texts in Mathematics. Springer-Verlag, New York. [42] Brieskorn, Egbert (1973). Sur les groupes de tresses [d’après V. I. Arnol d]. In Séminaire Bourbaki, 24ème année (1971/1972), Exp. No. 401, pp. 21–44. Lecture Notes in Math., Vol. 317. Springer, Berlin.

435

436

References

[43] Brown, Jr., Edgar H. and Szczarba, Robert H. (1997). On the rational homotopy type of function spaces. Trans. Amer. Math. Soc., 349(12), 4931–4951. [44] Brylinski, Jean-Luc (1988). A differential complex for Poisson manifolds. J. Differential Geom., 28(1), 93–114. [45] Buchstaber, Victor M. and Panov, Taras E. (2002). Torus Actions and their Applications in Topology and Combinatorics, Volume 24 of University Lecture Series. American Mathematical Society, Providence, RI. [46] Buijs, Urtzi and Murillo, Aniceto (2006). Basic constructions in rational homotopy theory of function spaces. Ann. Inst. Fourier (Grenoble), 56(3), 815–838. [47] Burghelea, Dan and Fiedorowicz, Zbigniew (1986). Cyclic homology and algebraic K-theory of spaces. II. Topology, 25(3), 303–317. [48] Cairns, Grant and Jessup, Barry (1997). New bounds on the Betti numbers of nilpotent Lie algebras. Comm. Algebra, 25(2), 415–430. [49] Calabi, Eugenio (1958). Construction and properties of some 6-dimensional almost complex manifolds. Trans. Amer. Math. Soc., 87, 407–438. [50] Cartan, Élie (1936). La topologie des espaces représentatifs des groupes de Lie, Volume 358 of Actualités Scientifiques et Industrielles. Hermann, Paris. [51] Cavalcanti, Gil (2007). The Lefschetz property, formality and blowing up in symplectic geometry. Trans. Amer. Math. Soc., 359 no. 1, 333–348. [52] Cavalcanti, Gil R. (2004). New aspects of the dd c -lemma. PhD thesis, University of Oxford, arXiv:math.DG/0501406. [53] Chas, Moira and Sullivan, Dennis (1999). String topology. arXiv:math. GT/9911159. [54] Cheeger, Jeff (1973). Some examples of manifolds of nonnegative curvature. J. Differential Geometry, 8, 623–628. [55] Cheeger, Jeff and Gromoll, Detlef (1971/72). The splitting theorem for manifolds of nonnegative Ricci curvature. J. Differential Geometry, 6, 119–128. [56] Cheeger, Jeff and Gromoll, Detlef (1972). On the structure of complete manifolds of nonnegative curvature. Ann. of Math. (2), 96, 413–443. [57] Chen, Kuo-tsai (1971/72). Differential forms and homotopy groups. J. Differential Geometry, 6, 231–246. [58] Chen, Kuo Tsai (1976). Reduced bar constructions on de Rham complexes. In Algebra, Topology, and Category Theory (A Collection of

References

[59] [60]

[61]

[62] [63]

[64]

[65]

[66]

[67]

[68] [69] [70] [71]

[72]

Papers in Honor of Samuel Eilenberg), pp. 19–32. Academic Press, New York. Chen, Kuo Tsai (1977). Iterated path integrals. Bull. Amer. Math. Soc., 83(5), 831–879. Chevalley, Claude (1999). Theory of Lie Groups. I, Volume 8 of Princeton Mathematical Series. Princeton University Press, Princeton, NJ. Fifteenth printing, Princeton Landmarks in Mathematics. Cohen, Frederick R., Lada, Thomas J., and May, J. Peter (1976). The Homology of Iterated Loop Spaces. Springer-Verlag, Berlin. Lecture Notes in Mathematics, Vol. 533. Cohen, Ralph L. and Jones, John D. S. (2002). A homotopy theoretic realization of string topology. Math. Ann., 324(4), 773–798. Conner, P. E. and Raymond, Frank (1970). Actions of compact Lie groups on aspherical manifolds. In Topology of Manifolds (Proc. Inst., Univ. of Georgia, Athens, Ga., 1969), pp. 227–264. Markham, Chicago, Ill. Cordero, Luis A., Fernández, Marisa, and Gray, Alfred (1986). Symplectic manifolds with no Kähler structure. Topology, 25(3), 375–380. Cordero, Luis A., Fernández, Marisa, and Gray, Alfred (1987). The Frölicher spectral sequence and complex compact nilmanifolds. C. R. Acad. Sci. Paris Sér. I Math., 305(17), 753–756. Cornea, Octav, Lupton, Gregory, Oprea, John, and Tanré, Daniel (2003). Lusternik–Schnirelmann Category, Volume 103 of Mathematical Surveys and Monographs. American Mathematical Society, Providence, RI. Datta, Basudeb and Subramanian, Swaminathan (1990). Nonexistence of almost complex structures on products of even-dimensional spheres. Topology Appl., 36(1), 39–42. Debongnie, Gery (2007). Rational homotopy type of subspace arrangements. arXiv:0705.1449v1 [math.AT]. Deligne, Pierre (1971). Théorie de Hodge. II. Inst. Hautes Études Sci. Publ. Math., 40, 5–57. Deligne, Pierre (1974). Théorie de Hodge. III. Inst. Hautes Études Sci. Publ. Math., 44, 5–77. Deligne, Pierre, Griffiths, Phillip, Morgan, John, and Sullivan, Dennis (1975). Real homotopy theory of Kähler manifolds. Invent. Math., 29(3), 245–274. Denham, Graham and Suciu, Alexander I. (2007). Moment-angle complexes, monomial ideals, and Massey products. Pure Appl. Math. Q., 3, 25 – 60.

437

438

References

[73] Deninger, Christopher and Singhof, Wilhelm (1988). On the cohomology of nilpotent Lie algebras. Bull. Soc. Math. France, 116(1), 3–14. [74] Dold, Albrecht (1963). Partitions of unity in the theory of fibrations. Ann. of Math. (2), 78, 223–255. [75] Dwyer, William G. and Wilkerson, Clarence W., Jr. (1998). The elementary geometric structure of compact Lie groups. Bull. London Math. Soc., 30(4), 337–364. [76] Eckmann, Beno and Frölicher, Alfred (1951). Sur l’intégrabilité des structures presque complexes. C. R. Acad. Sci. Paris, 232, 2284–2286. [77] Eells, Jr., James (1966). A setting for global analysis. Bull. Amer. Math. Soc., 72, 751–807. [78] Ehresmann, Charles (1952). Sur les variétés presque complexes. In Proceedings of the International Congress of Mathematicians, Cambridge, Mass., 1950, vol. 2, Providence, R. I., pp. 412–419. Amer. Math. Soc. [79] Ehresmann, Charles and Libermann, Paulette (1951). Sur les structures presque hermitiennes isotropes. C. R. Acad. Sci. Paris, 232, 1281–1283. [80] Ekedahl, Torsten (1986). Two examples of smooth projective varieties with nonzero Massey products. In Algebra, Algebraic Topology and their Interactions (Stockholm, 1983), Volume 1183 of Lecture Notes in Math., pp. 128–132. Springer, Berlin. [81] Eschenburg, Jost-Hinrich (1982). New examples of manifolds with strictly positive curvature. Invent. Math., 66(3), 469–480. [82] Eschenburg, Jost-Hinrich (1992). Cohomology of biquotients. Manuscripta Math., 75(2), 151–166. [83] Fadell, Edward and Neuwirth, Lee (1962). Configuration spaces. Math. Scand., 10, 111–118. [84] Fang, Fuquan and Rong, Xiaochun (2001). Curvature, diameter, homotopy groups, and cohomology rings. Duke Math. J., 107(1), 135–158. [85] Feichtner, Eva Maria and Yuzvinsky, Sergey (2005). Formality of the complements of subspace arrangements with geometric lattices. Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI), 326 (Teor. Predst. Din. Sist. Komb. i Algoritm. Metody. 13), 235–247, 284. [86] Félix, Yves and Halperin, Stephen (1982). Rational LS category and its applications. Trans. Amer. Math. Soc., 273(1), 1–38.

References

[87] Félix, Yves, Halperin, Stephen, and Thomas, Jean-Claude (2001). Rational Homotopy Theory, Volume 205 of Graduate Texts in Mathematics. Springer-Verlag, New York. [88] Félix, Yves and Tanré, Daniel (2005). The cohomology algebra of unordered configuration spaces. J. London Math. Soc. (2), 72(2), 525–544. [89] Félix, Yves and Tanré, Daniel (2005). H-space structure on pointed mapping spaces. Algebr. Geom. Topol., 5, 713–724 (electronic). [90] Félix, Yves and Thomas, Jean-Claude (1994). Homologie des espaces de lacets des espaces de configuration. Ann. Inst. Fourier (Grenoble), 44(2), 559–568. [91] Félix, Yves and Thomas, Jean-Claude (2000). Rational Betti numbers of configuration spaces. Topology Appl., 102(2), 139–149. [92] Félix, Yves and Thomas, Jean-Claude (2004). Configuration spaces and Massey products. Int. Math. Res. Not., 33, 1685–1702. [93] Félix, Yves, Thomas, Jean-Claude, and Vigué-Poirrier, Micheline (2004). The Hochschild cohomology of a closed manifold. Publ. Math. Inst. Hautes Études Sci., 99, 235–252. [94] Félix, Yves, Thomas, Jean-Claude, and Vigué-Poirrier, Micheline (2007). Rational string topology. J. Eur. Math. Soc. (JEMS), 9(1), 123–156. [95] Fernández, Marisa, de León, Manuel, and Saralegui, Martín (1996). A six-dimensional compact symplectic solvmanifold without Kähler structures. Osaka J. Math., 33(1), 19–35. [96] Fernández, Marisa and Gray, Alfred (1990). Compact symplectic solvmanifolds not admitting complex structures. Geom. Dedicata, 34(3), 295–299. [97] Fernández, Marisa and Muñoz, Vicente (2005). Formality of Donaldson submanifolds. Math. Z., 250(1), 149–175. [98] Fine, Benjamin L. and Triantafillou, Georgia (1993). On the equivariant formality of Kähler manifolds with finite group action. Canad. J. Math., 45(6), 1200–1210. [99] Franks, John (1992). Geodesics on S2 and periodic points of annulus homeomorphisms. Invent. Math., 108(2), 403–418. [100] Frauenfelder, Urs and Schlenk, Felix (2006). Fiberwise volume growth via Lagrangian intersections. J. Symplectic Geom., 4(2), 117–148. [101] Friedlander, John and Halperin, Stephen (1979). Distinct representatives, varieties and rational homotopy. J. Number Theory, 11 (3 S. Chowla Anniversary Issue), 321–323. [102] Frölicher, Alfred (1955). Zur Differentialgeometrie der komplexen Strukturen. Math. Ann., 129, 50–95.

439

440

References

[103] Fulton, William and Harris, Joe (1991). Representation Theory, Volume 129 of Graduate Texts in Mathematics. Springer-Verlag, New York. [104] Gallot, Sylvestre, Hulin, Dominique, and Lafontaine, Jacques (1987). Riemannian Geometry. Universitext. Springer-Verlag, Berlin. [105] Getzler, Ezra, Jones, John D. S., and Petrack, Scott (1991). Differential forms on loop spaces and the cyclic bar complex. Topology, 30(3), 339–371. [106] Godbillon, Claude (1974). Cohomologies d’algèbres de Lie de champs de vecteurs formels. In Séminaire Bourbaki, 25ème année (1972/1973), Exp. No. 421, pp. 69–87. Lecture Notes in Math., Vol. 383. Springer, Berlin. [107] Goldberg, Samuel I. (1962). Curvature and Homology. Pure and Applied Mathematics, Vol. XI. Academic Press, New York. [108] Gompf, Robert E. (1995). A new construction of symplectic manifolds. Ann. of Math. (2), 142(3), 527–595. [109] Goodwillie, Thomas G. (1985). Cyclic homology, derivations, and the free loopspace. Topology, 24(2), 187–215. [110] Goresky, Mark and Hingston, Nancy (2007). Loop products and closed geodesics. arXiv:0707.3486v [math.AT]. [111] Goyo, John O. (1989). The Sullivan model of the homotopy-fixedpoint set. PhD thesis, University of Toronto. [112] Greub, Werner, Halperin, Stephen, and Vanstone, Ray (1972). Connections, Curvature, and Cohomology. Vol. I: De Rham Cohomology of Manifolds and Vector Bundles. Academic Press, New York. Pure and Applied Mathematics, Vol. 47-49. [113] Greub, Werner, Halperin, Stephen, and Vanstone, Ray (1973). Connections, Curvature, and Cohomology. Vol. II: Lie Groups, Principal Bundles, and Characteristic Classes. Academic Press [Harcourt Brace Jovanovich, Publishers], New York. Pure and Applied Mathematics, Vol. 47-II. [114] Greub, Werner, Halperin, Stephen, and Vanstone, Ray (1976). Connections, Curvature, and Cohomology. Vol. III: Cohomology of Principal Bundles and Homogeneous Spaces. Academic Press [Harcourt Brace Jovanovich Publishers], New York. Pure and Applied Mathematics, Vol. 47-III. [115] Griffiths, Phillip and Harris, Joseph (1978). Principles of Algebraic Geometry. Wiley-Interscience [John Wiley & Sons], New York. Pure and Applied Mathematics. [116] Griffiths, Phillip A. and Morgan, John W. (1981). Rational Homotopy Theory and Differential Forms, Volume 16 of Progress in Mathematics. Birkhäuser Boston, Mass.

References

[117] Gromoll, Detlef and Meyer, Wolfgang (1969). Periodic geodesics on compact riemannian manifolds. J. Differential Geometry, 3, 493–510. [118] Gromoll, Detlef and Meyer, Wolfgang (1974). An exotic sphere with nonnegative sectional curvature. Ann. of Math. (2), 100, 401–406. [119] Gromov, Mikhael (1978). Homotopical effects of dilatation. J. Differential Geom., 13(3), 303–310. [120] Grove, Karsten (1973). Condition (C) for the energy integral on certain path spaces and applications to the theory of geodesics. J. Differential Geometry, 8, 207–223. [121] Grove, Karsten (1974). Isometry-invariant geodesics. Topology, 13, 281–292. [122] Grove, Karsten and Halperin, Stephen (1982). Contributions of rational homotopy theory to global problems in geometry. Inst. Hautes Études Sci. Publ. Math., 56, 171–177 (1983). [123] Grove, Karsten, Halperin, Stephen, and Vigué-Poirrier, Micheline (1978). The rational homotopy theory of certain path spaces with applications to geodesics. Acta Math., 140(3-4), 277–303. [124] Gualtieri, Marco (2003). Generalized complex geometry. PhD thesis, University of Oxford, arXiv:math.DG/0401221. [125] Haefliger, André (1976). Sur la cohomologie de l’algèbre de Lie des champs de vecteurs. Ann. Sci. École Norm. Sup. (4), 9(4), 503–532. [126] Haefliger, André (1982). Rational homotopy of the space of sections of a nilpotent bundle. Trans. Amer. Math. Soc., 273(2), 609–620. [127] Hain, Richard M. (1984). Iterated integrals and homotopy periods. Mem. Amer. Math. Soc., 47(291), iv+98. [128] Hain, Richard M. (1987). The de Rham homotopy theory of complex algebraic varieties. I. K-Theory, 1(3), 271–324. [129] Halperin, Stephen (1977). Finiteness in the minimal models of Sullivan. Trans. Amer. Math. Soc., 230, 173–199. [130] Halperin, Stephen (1983). Lectures on minimal models. Mém. Soc. Math. France (N.S.), 9-10, 261. [131] Halperin, Stephen (1985). Rational homotopy and torus actions. In Aspects of Topology, Volume 93 of London Math. Soc. Lecture Note Ser., Cambridge, pp. 293–306. Cambridge Univ. Press. [132] Halperin, Stephen and Stasheff, James (1979). Obstructions to homotopy equivalences. Adv. in Math., 32(3), 233–279. [133] Halperin, Stephen and Tanré, Daniel (1990). Homotopie filtrée et fibrés C ∞ . Illinois J. Math., 34(2), 284–324. [134] Hasegawa, Keizo (1989). Minimal models of nilmanifolds. Proc. Amer. Math. Soc., 106(1), 65–71.

441

442

References

[135] Helgason, Sigur¯dur (1962). Differential Geometry and Symmetric Spaces. Pure and Applied Mathematics, Vol. XII. Academic Press, New York. [136] Helgason, Sigur¯dur (2001). Differential Geometry, Lie Groups, and Symmetric Spaces, Volume 34 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI. Corrected reprint of the 1978 original. [137] Hilton, Peter (1965). Homotopy Theory and Duality. Gordon and Breach Science Publishers, New York. [138] Hilton, Peter, Mislin, Guido, and Roitberg, Joe (1975). Localization of Nilpotent Groups and Spaces. North-Holland Publishing Co., Amsterdam. North-Holland Mathematics Studies, No. 15, Notas de Matemática, No. 55. [Notes on Mathematics, No. 55]. [139] Hilton, Peter and Stammbach, Urs (1997). A Course in Homological Algebra (Second edn), Volume 4 of Graduate Texts in Mathematics. Springer-Verlag, New York. [140] Hingston, Nancy (1993). On the growth of the number of closed geodesics on the two-sphere. Internat. Math. Res. Notices, 9, 253–262. [141] Hirzebruch, Friedrich (1966). Topological Methods in Algebraic Geometry. Third enlarged edition. New appendix and translation from the second German edition by R. L. E. Schwarzenberger, with an additional section by A. Borel. Die Grundlehren der Mathematischen Wissenschaften, Band 131. Springer-Verlag New York. [142] Hitchin, Nigel (2003). Generalized Calabi–Yau manifolds. Quarterly Journal Math., 54(3), 281–308. [143] Hopf, Heinz (1948). Zur Topologie der komplexen Mannigfaltigkeiten. In Studies and Essays Presented to R. Courant on his 60th Birthday, January 8, 1948, pp. 167–185. Interscience Publishers, Inc., New York. [144] Hsiang, Wu-Yi (1975). Cohomology Theory of Topological Transformation Groups. Springer-Verlag, New York. Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 85. [145] Husemoller, Dale (1994). Fiber Bundles (Third edn), Volume 20 of Graduate Texts in Mathematics. Springer-Verlag, New York. [146] Ibáñez, Raoul, K¸edra, Jarek, Rudyak, Yuli, and Tralle, Aleksy (2004). On fundamental groups of symplectically aspherical manifolds. Math. Z., 248(4), 805–826. [147] Iwasawa, Kenkichi (1949). On some types of topological groups. Ann. of Math. (2), 50, 507–558.

References

[148] Jessup, Barry and Lupton, Gregory (2004). Free torus actions and two-stage spaces. Math. Proc. Cambridge Philos. Soc., 137(1), 191–207. [149] Jones, John D. S. (1987). Cyclic homology and equivariant homology. Invent. Math., 87(2), 403–423. [150] Kapovitch, Vitali (2004). A note on rational homotopy of biquotients. Preprint. [151] Kapovitch, Vitali and Ziller, Wolfgang (2004). Biquotients with singly generated rational cohomology. Geom. Dedicata, 104, 149–160. [152] Kawakubo, Katsuo (1991). The Theory of Transformation Groups (Japanese edn). The Clarendon Press Oxford University Press, New York. [153] K¸edra, Jarosław (2000). KS-models and symplectic structures on total spaces of bundles. Bull. Belg. Math. Soc. Simon Stevin, 7(3), 377–385. [154] Klingenberg, Wilhelm (1978). Lectures on Closed Geodesics. Springer-Verlag, Berlin. Grundlehren der Mathematischen Wissenschaften, Vol. 230. [155] Kobayashi, Shoshichi and Nomizu, Katsumi (1969). Foundations of Differential Geometry. Vol. II. Interscience Tracts in Pure and Applied Mathematics, No. 15 Vol. II. Interscience Publishers John Wiley & Sons, Inc., New York. [156] Kontsevich, Maxim (1999). Operads and motives in deformation quantization. Lett. Math. Phys., 48(1), 35–72. Moshé Flato (1937–1998). [157] Koszul, Jean-Louis (1985). Crochet de Schouten-Nijenhuis et cohomologie. Astérisque, Numero Hors Serie, 257–271. The mathematical heritage of Élie Cartan (Lyon, 1984). [158] Kotani, Yasusuke (2004). Note on the rational cohomology of the function space of based maps. Homology Homotopy Appl., 6(1), 341–350 (electronic). [159] Kreck, M. and Schafer, J. A. (1984). Classification and stable classification of manifolds: some examples. Comment. Math. Helv., 59(1), 12–38. [160] Kriz, Igor (1994). On the rational homotopy type of configuration spaces. Ann. of Math. (2), 139(2), 227–237. [161] Lalonde, François and McDuff, Dusa (2003). Symplectic structures on fiber bundles. Topology, 42(2), 309–347. [162] Lalonde, François, McDuff, Dusa, and Polterovich, Leonid (1998). On the flux conjectures. In Geometry, Topology, and Dynamics

443

444

References

[163] [164]

[165] [166]

[167] [168]

[169]

[170]

[171]

[172] [173]

[174] [175] [176] [177] [178]

(Montreal, PQ, 1995), Volume 15 of CRM Proc. Lecture Notes, pp. 69–85. Amer. Math. Soc., Providence, RI. Lambre, Thierry (1991). Modèle minimal équivariant et formalité. Trans. Amer. Math. Soc., 327(2), 621–639. Lambrechts, Pascal (2001). The Betti numbers of the free loop space of a connected sum. J. London Math. Soc. (2), 64(1), 205–228. Lambrechts, Pascal and Stanley, Don (2004). The rational homotopy type of a blow-up in the stable case. arXiv:0706.1363v1 [math.AT]. Lambrechts, Pascal and Stanley, Don (2004). The rational homotopy type of configuration spaces of two points. Ann. Inst. Fourier (Grenoble), 54(4), 1029–1052. Lambrechts, Pascal and Stanley, Don (2005). Algebraic models of Poincaré embeddings. Algebr. Geom. Topol., 5, 135–182. Lambrechts, Pascal and Stanley, Don (2005). Examples of rational homotopy types of blow-ups. Proc. Amer. Math. Soc., 133(12), 3713–3719. Lambrechts, Pascal and Stanley, Don (2005). Poincaré duality and commutative differential graded algebras. arXiv:math/0701309v2 [math.AT]. Lehmann, Daniel (1977). Théorie homotopique des formes différentielles (d’après D. Sullivan). Société Mathématique de France, Paris. Astérisque, No. 45. Lemaire, Jean-Michel (1974). Algèbres connexes et homologie des espaces de lacets. Springer-Verlag, Berlin. Lecture Notes in Mathematics, Vol. 422. Levitt, Norman (1995). Spaces of arcs and configuration spaces of manifolds. Topology, 34(1), 217–230. Libermann, Paulette (1954). Sur le problème d’équivalence de certaines structures infinitésimales. Ann. Mat. Pura Appl. (4), 36, 27–120. Lichnerowicz, André (1977). Les variétés de Poisson et leurs algèbres de Lie associées. J. Differential Geometry, 12(2), 253–300. Lin, Yi and Sjamaar, Reyer (2004). Equivariant symplectic Hodge theory and the dG δ-lemma. J. Symplectic Geom., 2(2), 267–278. Longoni, Riccardo and Salvatore, Paolo (2005). Configuration spaces are not homotopy invariant. Topology, 44(2), 375–380. Lupton, Gregory (1990). Intrinsic formality and certain types of algebras. Trans. Amer. Math. Soc., 319(1), 257–283. Lupton, Gregory and Oprea, John (1994). Symplectic manifolds and formality. J. Pure Appl. Algebra, 91(1-3), 193–207.

References

[179] Lupton, Gregory and Oprea, John (1995). Cohomologically symplectic spaces: toral actions and the Gottlieb group. Trans. Amer. Math. Soc., 347(1), 261–288. [180] Lupton, Gregory and Smith, Samuel B. (2007). Rationalized evaluation subgroups of a map i: Sullivan models, derivations and G-sequences. Journal of Pure and Applied Algebra, 209, 159–171. [181] Lyusternik, Lazar Aronovich and Fet, Abram Il’ich (1951). Variational problems on closed manifolds. Doklady Akad. Nauk SSSR (N.S.), 81, 17–18. [182] Mac Lane, Saunders (1995). Homology. Classics in Mathematics. Springer-Verlag, Berlin. Reprint of the 1975 edition. [183] Malcev, Anatoli Ivanovich (1962). On a class of homogeneous spaces. Amer. Math. Soc. Translations Series 1, 9, 276–307. [184] Mather, Michael (1976). Pull-backs in homotopy theory. Canad. J. Math., 28(2), 225–263. [185] Mathieu, Olivier (1995). Harmonic cohomology classes of symplectic manifolds. Comment. Math. Helv., 70(1), 1–9. [186] McCleary, John (1985). User’s Guide to Spectral Sequences, Volume 12 of Mathematics Lecture Series. Publish or Perish Inc., Wilmington, DE. [187] McDuff, Dusa (1975). Configuration spaces of positive and negative particles. Topology, 14, 91–107. [188] McDuff, Dusa (1984). Examples of simply-connected symplectic non-Kählerian manifolds. J. Differential Geom., 20(1), 267–277. [189] McDuff, Dusa and Salamon, Dietmar (1998). Introduction to Symplectic Topology (Second edn). Oxford Mathematical Monographs. The Clarendon Press Oxford University Press, New York. [190] Merkulov, Sergei (1998). Formality of canonical symplectic complexes and Frobenius manifolds. Internat. Math. Res. Notices, 14, 727–733. [191] Merkulov, Sergei (2004). De Rham model for string topology. Int. Math. Res. Not., 55, 2955–2981. [192] Miller, Timothy James (1979). On the formality of (k−1)-connected compact manifolds of dimension less than or equal to 4k − 2. Illinois J. Math., 23(2), 253–258. [193] Milnor, John (1956). Construction of universal bundles. II. Ann. of Math. (2), 63, 430–436. [194] Milnor, John (1958). On the existence of a connection with curvature zero. Comment. Math. Helv., 32, 215–223.

445

446

References

[195] Milnor, John (1963). Morse Theory. Based on lecture notes by M. Spivak and R. Wells. Annals of Mathematics Studies, No. 51. Princeton University Press, Princeton, N.J. [196] Milnor, John (1968). A note on curvature and fundamental group. J. Differential Geometry, 2, 1–7. [197] Milnor, John W. and Moore, John C. (1965). On the structure of Hopf algebras. Ann. of Math. (2), 81, 211–264. [198] Milnor, John W. and Stasheff, James D. (1974). Characteristic Classes. Princeton University Press, Princeton, N. J. Annals of Mathematics Studies, No. 76. [199] Mimura, Mamoru and Toda, Hirosi (1991). Topology of Lie Groups. I, II, Volume 91 of Translations of Mathematical Monographs. American Mathematical Society, Providence, RI. Translated from the 1978 Japanese edition by the authors. [200] Moore, John (1954-55). Exposés 22. Séminaire Cartan, Paris. [201] Morgan, John W. (1978). The algebraic topology of smooth algebraic varieties. Inst. Hautes Études Sci. Publ. Math., 48, 137–204. [202] Morgan, John W. (1986). Correction to: “The algebraic topology of smooth algebraic varieties” [Inst. Hautes Études Sci. Publ. Math. No. 48 (1978), 137–204; MR0516917 (80e:55020)]. Inst. Hautes Études Sci. Publ. Math., 64, 185. [203] Mo/ller, Jesper Michael (1987). Nilpotent spaces of sections. Trans. Amer. Math. Soc., 303(2), 733–741. [204] Mo/ller, Jesper Michael (2002). Rational equivalences between classifying spaces. Math. Z., 241(4), 761–799. [205] Myers, Sumner Byron and Steenrod, Norman E. (1939). The group of isometries of a Riemannian manifold. Ann. of Math. (2), 40(2), 400–416. [206] Neisendorfer, Joseph and Miller, Timothy (1978). Formal and coformal spaces. Illinois J. Math., 22(4), 565–580. [207] Neisendorfer, Joseph and Taylor, Laurence (1978). Dolbeault homotopy theory. Trans. Amer. Math. Soc., 245, 183–210. [208] Newlander, August and Nirenberg, Louis (1957). Complex analytic coordinates in almost complex manifolds. Ann. of Math. (2), 65, 391–404. [209] Nomizu, Katsumi (1954). On the cohomology of compact homogeneous spaces of nilpotent Lie groups. Ann. of Math. (2), 59, 531–538. [210] Notbohm, Dietrich (1995). Classifying spaces of compact Lie groups and finite loop spaces. In Handbook of Algebraic Topology, pp. 1049–1094. North-Holland, Amsterdam.

References

[211] Notbohm, Dietrich (1995). On the “classifying space” functor for compact Lie groups. J. London Math. Soc. (2), 52(1), 185–198. [212] Notbohm, Dietrich and Ray, Nigel (2005). On Davis–Januszkiewicz homotopy types. I. Formality and rationalisation. Algebr. Geom. Topol., 5, 31–51. [213] O’Neill, Barrett (1966). The fundamental equations of a submersion. Michigan Math. J., 13, 459–469. [214] Onishchik, Arkady L. and Vinberg, Èrnest B. (1993). Foundations of Lie theory. In Lie Groups and Lie Algebras, I, Volume 20 of Encyclopaedia Math. Sci., pp. 1–94, 231–235. Springer, Berlin. [215] Oprea, John (1984). Lifting homotopy actions in rational homotopy theory. J. Pure Appl. Algebra, 32(2), 177–190. [216] Oprea, John (2004). Differential Geometry and Its Applications, 2nd edn. Prentice Hall, England U.K. [217] Oprea, John and Tanré, Daniel (2005). Flat circle bundles, pullbacks, and the circle made discrete. Int. J. Math. Math. Sci., 21, 3487–3495. [218] Osse, Akimou (1992). Sur le classifiant d’un groupe de Lie compact connexe. C. R. Acad. Sci. Paris Sér. I Math., 315(7), 833–838. [219] Özaydin, Murad and Walschap, Gerard (1994). Vector bundles with no soul. Proc. Amer. Math. Soc., 120(2), 565–567. [220] Pan, Jian Zhong and Wu, Shao Bing (2006). Rational homotopy theory and nonnegative curvature. Acta Math. Sin. (Engl. Ser.), 22(1), 23–26. [221] Papadima, Stefan ¸ (1986). Rigidity properties of compact Lie groups modulo maximal tori. Math. Ann., 275(4), 637–652. [222] Papadima, Stefan ¸ (1988). Rational homotopy equivalences of Lie type. Math. Proc. Cambridge Philos. Soc., 104(1), 65–80. [223] Paternain, Gabriel P. (1992). On the topology of manifolds with completely integrable geodesic flows. Ergodic Theory Dynam. Systems, 12(1), 109–121. [224] Paternain, Gabriel P. (1999). Geodesic Flows, Volume 180 of Progress in Mathematics. Birkhäuser Boston Inc., Boston, MA. [225] Paternain, Gabriel P. and Paternain, Miguel (1994). Topological entropy versus geodesic entropy. Internat. J. Math., 5(2), 213–218. [226] Petersen, Peter (1998). Riemannian Geometry, Volume 171 of Graduate Texts in Mathematics. Springer-Verlag, New York. [227] Pittie, Harsh V. (1988). The Dolbeault-cohomology ring of a compact, even- dimensional Lie group. Proc. Indian Acad. Sci. Math. Sci., 98(2-3), 117–152. [228] Pittie, Harsh V. (1989). The nondegeneration of the Hodge-de Rham spectral sequence. Bull. Amer. Math. Soc. (N.S.), 20(1), 19–22.

447

448

References

[229] Postnikov, Mikhail Mikhailovich (1990). Leçons de géométrie. Traduit du Russe: Mathématiques. [Translations of Russian Works: Mathematics]. Éditions Mir, Moscow. Géométrie différentielle. [Differential geometry], Translated from the Russian by Irina Petrova. [230] Rudyak, Yuli and Tralle, Aleksy (2000). On Thom spaces, Massey products, and nonformal symplectic manifolds. Internat. Math. Res. Notices, 10, 495–513. [231] Scheerer, Hans (1968). Homotopieäquivalente kompakte Liesche Gruppen. Topology, 7, 227–232. [232] Scheerer, Hans and Tanré, Daniel (1992). Homotopie modérée et tempérée avec les coalgèbres. Applications aux espaces fonctionnels. Arch. Math. (Basel), 59(2), 130–145. [233] Schultz, Reinhard (1971). The nonexistence of free S1 actions on some homotopy spheres. Proc. Amer. Math. Soc., 27, 595–597. [234] Scorpan, Alexandru (2005). The Wild World of 4-Manifolds. American Mathematical Society, Providence, RI. [235] Scull, Laura (2004). Equivariant formality for actions of torus groups. Canad. J. Math., 56(6), 1290–1307. [236] Serre, Jean-Pierre (1951). Homologie singulière des espaces fibrés. Applications. Ann. of Math. (2), 54, 425–505. [237] Shiga, Hiroo and Tezuka, Michishige (1987). Rational fibrations, homogeneous spaces with positive Euler characteristics and Jacobians. Ann. Inst. Fourier (Grenoble), 37(1), 81–106. [238] Singhof, Wilhelm (1993). On the topology of double coset manifolds. Math. Ann., 297(1), 133–146. [239] Smale, Stephen (1959). The classification of immersions of spheres in Euclidean spaces. Ann. of Math. (2), 69, 327–344. [240] Spanier, Edwin H. (1966). Algebraic Topology. McGraw-Hill, New York. [241] Stasheff, James (1983). Rational Poincaré duality spaces. Illinois J. Math., 27(1), 104–109. [242] Steenrod, Norman (1999). The topology of fiber bundles. Princeton Landmarks in Mathematics. Princeton University Press, Princeton, NJ. Reprint of the 1957 edition, Princeton Paperbacks. [243] St¸epien, ´ Zofia (2006). The Lalonde–McDuff conjecture for nilmanifolds. arXiv:math/0608214v1 [math.SG]. [244] Stong, Robert E. (1965). Relations among characteristic numbers. I. Topology, 4, 267–281. [245] Sullivan, Dennis (1975). Differential forms and the topology of manifolds. In Manifolds—Tokyo 1973 (Proc. Internat. Conf., Tokyo, 1973), pp. 37–49. Univ. Tokyo Press, Tokyo.

References

[246] Sullivan, Dennis (1977). Infinitesimal computations in topology. Inst. Hautes Études Sci. Publ. Math., 47, 269–331 (1978). [247] Tanaka, Minoru (1982). On the existence of infinitely many isometry-invariant geodesics. J. Differential Geom., 17(2), 171–184. [248] Tanré, Daniel (1983). Homotopie rationnelle: modèles de Chen, Quillen, Sullivan, Volume 1025 of Lecture Notes in Mathematics. Springer-Verlag, Berlin. [249] Tanré, Daniel (1994). Modèle de Dolbeault et fibré holomorphe. J. Pure Appl. Algebra, 91(1-3), 333–345. [250] Thomas, Jean-Claude (1981). Rational homotopy of Serre fibrations. Ann. Inst. Fourier (Grenoble), 31(3), v, 71–90. [251] Tischler, David (1977). Closed 2-forms and an embedding theorem for symplectic manifolds. J. Differential Geometry, 12(2), 229–235. [252] tom Dieck, Tammo (1987). Transformation Groups, Volume 8 of de Gruyter Studies in Mathematics. Walter de Gruyter & Co., Berlin. [253] Toomer, Graham Hilton (1975). Topological localization, category and cocategory. Canad. J. Math., 27, 319–322. [254] Totaro, Burt (1996). Configuration spaces of algebraic varieties. Topology, 35(4), 1057–1067. [255] Totaro, Burt (2002). Cheeger manifolds and the classification of biquotients. J. Differential Geom., 61(3), 397–451. [256] Totaro, Burt (2003). Curvature, diameter, and quotient manifolds. Math. Res. Lett., 10(2-3), 191–203. [257] Tralle, Aleksy and Oprea, John (1997). Symplectic Manifolds with no Kähler Structure, Volume 1661 of Lecture Notes in Mathematics. Springer-Verlag, Berlin. [258] Triantafillou, Georgia (1978). Äquivariante rationale Homotopietheorie. Bonner Mathematische Schriften [Bonn Mathematical Publications], 110. Universität Bonn Mathematisches Institut, Bonn. Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn, Bonn, 1977. [259] Tuschmann, Wilderich (2002). Geometric diffeomorphism finiteness in low dimensions and homotopy group finiteness. Math. Ann., 322(2), 413–420. [260] Vigué-Poirrier, Micheline (1986). Cohomologie de l’espace des sections d’un fibré et cohomologie de Gelfand-Fuchs d’une variété. In Algebra, Algebraic Topology and their Interactions (Stockholm, 1983), Volume 1183 of Lecture Notes in Math., pp. 371–396. Springer, Berlin.

449

450

References

[261] Vigué-Poirrier, Micheline and Sullivan, Dennis (1976). The homology theory of the closed geodesic problem. J. Differential Geometry, 11(4), 633–644. [262] Warner, Frank W. (1971). Foundations of Differentiable Manifolds and Lie Groups. Scott, Foresman and Co., Glenview, Ill.-London. [263] Weingram, Stephen (1971). On the incompressibility of certain maps. Ann. of Math. (2), 93, 476–485. [264] Wells, Raymond O., Jr. (1980). Differential Analysis on Complex Manifolds (Second edn), Volume 65 of Graduate Texts in Mathematics. Springer-Verlag, New York. [265] Whitehead, George W. (1978). Elements of Homotopy Theory, Volume 61 of Graduate Texts in Mathematics. Springer-Verlag, New York. [266] Yan, Dong (1996). Hodge structure on symplectic manifolds. Adv. Math., 120(1), 143–154. [267] Yuzvinsky, Sergey (1999). Rational model of subspace complement on atomic complex. Publ. Inst. Math. (Beograd) (N.S.), 66(80), 157–164. Geometric combinatorics (Kotor, 1998). [268] Zeeman, Eric Christopher (1957). A proof of the comparison theorem for spectral sequences. Proc. Cambridge Philos. Soc., 53, 57–62.

Index (B(A, A, A), D), 234 AN , 90 ADR (P∗,∗ (M)), 383 ADR (P∗,a (M)), 383 ADR (Pfree (M)), 378 APL , 67 construction of, 70 DA , 328, 355 F(Rn , k), 354 are formal, 354 G(x), 274 Gx , 274 MG , 274 Sh(k, l), 381 TopN , 90 , 406 L, 399 M≤D κ≤sec≤λ (n), 242 Q[V], 59 P∗,∗ (M), 383 P∗,a (M), 383 Pfree (M), 377 Chen Pfree (M), 379 Chen P∗,∗ (M), 384 Chen Pfree loop (M), 385 δ, 405 λ-derivation, 307 ∧V, 58, 59 CM, 141 rk0 , 272 d1 , 77 dk , 77 nT , 223 Bl(A, B), 336 abelian Lie algebra, 8 Lie group, 8 action, 273 almost free, 271, 274, 316 cohomologically free, 308 conjugation, 7 effective, 11

fixed point of, 274 free, 11, 274 Hamiltonian, 308 of π1 on πn , 31 of Lie group, 11 orbit of, 274 smooth, 274 space of orbits of, 274 symplectic, 306 adjoint representation, 5 algebra commutative differential graded, 58 commutative graded, 58 differential graded, 58 algebraic mapping cone, 322 almost complex manifold, 150 integrable, 154 map, 151 structure, 150 obstructions to, 156 on S2 and S6 , 151, 156 arrangement, 358 complement of, 358 formality of, 361 geometric, 358, 363 atomic complex relative, 359 augmented cdga, 60 bar construction, 236, 385 normalization, 382 double, 234, 381 Barratt–Puppe sequence, 430 basepoint nondegenerate, 429 basic complex, 48 bi-invariant form, 12 volume form, 13 bigradation, 92, 95, 169, 198 on C-forms, 153 bigraded differential algebra, 412

452

Index

model, 95 and Stasheff’s theorem, 113 biquotient, 134, 136 as homotopy pullback, 135 minimal model of, 138 model of, 137 non-negative curvature of, 242, 252 splitting rigid, 260 symplectic, 189 blow-up along a submanifold, 333 complex, 331 example of model, 334 McDuff’s example, 337 model for, 336, 337 properties of, 333 topological, 332 and hard Lefschetz, 339 Borel fibration, 47, 272, 275 relative model of, 280 fixed point criterion, 293 localization theorem, 272, 292 spectral sequence, 173 bracket, 398 of vector fields, 398, 408 Brylinski conjecture, 184, 339 counterexample, 185 bumpy metric, 224 bundle along the fibers, 340 holomorphic, 150 holomorphic vector, 150 principal holomorphic, 150 c-split fibration, 165, 313 Calabi–Eckmann manifold, 159, 160 canonical minimal model, 140, 141 of an isometry, 142 Cartan decomposition, 26 Cartan–Weil model, 51 Cartan–Weil model, 309 cdga, 58 augmented, 60 cohomologically symplectic, 188 filtered, 171 formal, 92 homotopy, 61 minimal, 59 pure, 419 r-bigraded, 170 Sullivan, 59

center, 7 of rational homotopy Lie algebra, 287 centralizer, 7 cga, 58 free, 58 characteristic algebra, 260 characteristic class, 46 Chas–Sullivan loop product, 342 Chern class, 46 classifying space, 38 cohomology of, 38, 43, 45 minimal model of, 72 closed form, 402, 406 geodesic, 205 as critical point, 208 problem, 207 manifold, 227 co-closed form, 405, 406 co-differential as adjoint of d, 405 on forms, 405 cofibration, 429 cofiber, 429 cofiber sequence, 430 cohomological shriek map, 319 cohomologically free action, 308 cohomologically symplectic cdga, 188 manifold, 188 cohomology de Rham, 402 equivariant, 276 Hochschild, 225, 237, 346 invariant, 125 of free loop space, 236, 385 of Lie algebra, 54 singular, 403 commutative differential graded algebra, 58 commutative graded algebra, 58 free, 58 complement of an arrangement, 358 model of, 360 complement of submanifold model of, 324 complex blow-up, 331 structure, 196 generalized, 202 on tangent space, 150 manifold, 148 torus, 149 complex projective space, 149

Index

complexification, 153, 197 configuration space, 351 Fadell–Neuwirth fibration, 352 formality of, 355 model of complex projective manifold, 355 model conjecture, 356 model of, 356 of two points, 329 rational homotopy of, 353 with labels, 374 conjecture Bott, 265, 390 fibration with torus fiber, 388 Halperin, 168, 389 on configuration space model, 356 toral rank, 271, 283, 388 conjugate point, 229, 232, 237 multiplicity of, 229 conjugation action, 7 connected sum, 107 and orientation of fundamental class, 108 is formal if summands are, 115 minimal model of, 108 rational homotopy of, 107 cotangent bundle as symplectic manifold, 183 critical value of a homology class, 348 cup length, 88 curvature Ricci, 269 Riemann, 240 scalar, 269 and Grassmann manifold, 240 sectional, 240 de Rham cohomology, 402 forms, 392 theorem, 403 decomposable elements, 59 dga, 58 diagonal class, 355 diagonal cocycle, 328 and shriek map, 329 dichotomy theorem elliptic case, 85 hyperbolic case, 85 differential decomposition of, 77 perturbation of, 96 differential forms, 392

differential graded algebra, 58 morphism, 58 dimension, 84 Dolbeault cohomology, 155 of S1 × S2n+1 , 174 of S2m+1 × S2n+1 , 177 complex, 155 formality, 169, 180 model, 169 of CP(n), 173 of Calabi–Eckmann manifolds, 176 double bar construction, 381 double complex, 420 double mapping cylinder, 426 Eilenberg–Mac Lane space, 87 and non-positive curvature, 241 elliptic, 84 4-manifolds, 108 energy, 208, 229 entropy, 227 equivalence n-, 423 weak, 423 equivariant cohomology, 47, 51, 276 equivariant minimal model, 123, 221 equivariantly formal, 127, 168 Euler characteristic homotopy, 85 of fixed point set, 128, 292 Euler class, 46, 257 and flat, 268 exponential, 6 extension of cdga homotopies, 64 exterior derivative, 395, 400 Fadell–Neuwirth fibration, 352 fiber bundle, 26 along the fibers, 340 exterior algebra, 398 pullback of, 27 symplectic, 190 tangent, 393 fibration, 79, 430 c-split, 313 Hamitonian, 165 holonomy of, 80 long exact homotopy sequence of, 79, 81 nilpotent, 79 path, 79, 431

453

454

Index

relative model of, 81 pure, 189 quasi-nilpotent, 79 relative minimal model of, 81 TNCZ, 164, 313 orientable, 164 fiber, 430 fiber bundle, 431 Hamiltonian, 314 fiber sequence, 431 filtered cdga, 171 differential graded algebra, 412 differential module, 412 model, 96 filtration, 412 bounded, 412 fixed point, 274 flat and vanishing Euler class, 268 connection, 255 manifold, 241, 256 principal bundle, 254 vector bundle, 255 flow associated to a vector field, 399, 408 formal cdga, 92 criterion for, 102 obstructions to, 97 formal homology connection, 386 of CP(2), 387 formal space, 92 F(Rn , k), 354 and connected sums, 115 criterion for, 99 Dolbeault, 169 equivariant, 126 Kähler, 166 Kähler manifold, 365 nonformal example, 94 over Q-extensions, 100 product of, 93 retract of, 93 Riemannian symmetric, 144 strict, 172 vanishing of all Massey products, 96 vanishing of triple Massey products, 94 forms de Rham, 392 Sullivan, 67 Four Lemma, 418 Frölicher spectral sequence, 179 of Kähler manifold, 179 free ccga, 58

free loop space cohomology, 385 model, 210, 369 fundamental 2-form, 157 fundamental vector field, 306 Gelfand–Fuchs cohomology, 375 geodesic and Morse theory, 229 closed, 205 flow, 227 completely integrable, 230 geometrically distinct, 205 index of, 229, 232 invariant, 215 as critical point, 216 on compact Lie group, 241 totally, 252 geometric lattice, 358 graded Lie algebra, 77 cochains on, 91 realization of, 92 Grassmann manifold, 43, 177 group action, 11, 273 almost free, 271, 274 and Euler characteristic, 128, 292 cohomologically free, 308 effective, 11 free, 11, 274 Hamiltonian, 308 orbit of, 274 smooth, 274 symplectic, 306 Grove’s question, 242 Gysin sequence, 44, 415 Haefliger model for evaluation map, 369 for mapping space, 369 Halperin conjecture, 168 Hamiltonian action, 308 criterion for, 310 fixed point set of, 310 is c-split, 313 fiber bundle, 314 with nilmanifold fiber, 315 hard Lefschetz and symplectically harmonic, 185 and TRC, 285 for blow-up, 339 property, 163, 165 theorem, 163

Index

harmonic form, 25, 158, 407 Heisenberg group, 118 manifold, 117, 162 hermitian product, 198 metric, 156 Hochschild cohomology, 237, 346 complex, 235, 385 homology, 236 Hodge decomposition, 139, 407 laplacian, 406 manifold, 160 star operator, 404 structure mixed, 366 pure, 364 holomorphic form, 155 map, 148, 149 holonomy, 432 representation, 80 homogeneous space, 29, 136 model of, 83 symplectic, 189 homotopy cofiber, 429 equivalence, 424 Euler characteristic, 85, 278 extension, 64 fiber, 431 Lie algebra, 106 lifting, 63 property, 431 of cdga morphisms, 61 left, 62 pullback, 424, 425 versus pullback, 432 pushout, 426, 427 versus pushout, 430 real, 69 type real, 143 Hopf fibration relative model of, 82 manifold, 159, 161, 174 surface, 161 Hopf’s theorem, 18 Hurewicz image, 306 hyperbolic, 84

indecomposables, 60 induced map on, 76 independent sequence, 360 index of a geodesic, 229 inner product on forms, 404 integrable almost complex manifold, 154 integration along a fiber, 160, 378, 379 of forms, 401 interior multiplication, 398 intersection map, 348 intersection product, 341 invariant cohomology, 13 invariant form, 12 L(X), 399 bi-invariant, 12 left, 12 invariant geodesic, 215 isometry, 215 finite order, 217 isotropy subgroup, 274 iterated integral, 377, 378, 385 product of, 381 Iwasawa decomposition, 3 Jacobi identity, 398, 408 k-invariants, 88 Kähler criteria for, 158 form, 157 manifold, 157, 166, 172 formality of, 166, 172, 365 metric, 157 Killing form, 25, 270 Kodaira–Thurston manifold, 156, 162, 165, 190 Chern classes of, 340 Lalonde–McDuff question, 314, 391 Laplace operator, 158 lattice geometric, 358 Lefschetz type, 192, 307 and TRC, 312 nilmanifold, 193 left adjoint of cdga tensor product, 371 left homotopy, 62 left invariant form, 12

455

456

Index

lemma d δ, 186 level-nilpotent, 348 Lie algebra, 3 2-step, 289 abelian, 8 cohomology, 54 free graded, 78, 92, 386 free product, 78 graded, 77 homomorphism, 4 homotopy, 106 nilpotent, 117 TRC, 289 Lie bracket, 3, 77, 408 Lie derivative, 399 and symplectic action, 307 formula, 400 Lie group, 2 abelian, 8 action, 11 classifying space of, 38 cohomology Hopf’s theorem, 18 cohomology of, 16, 43, 45 connection on, 270 even dimensional as complex manifold, 162 exponential, 6, 11 homomorphism, 2 Killing form of, 270 left-right translations, 4 made discrete, 266 maximal compact subgroup of, 3 minimal model of, 71 one-parameter subgroup of, 5 rank of, 8 second Betti number, 22, 40 sectional curvature of, 270 semisimple, 21, 270 simple, 21, 26 third Betti number, 23 Lie subalgebra, 4 Lie subgroup, 3 lifting lemma, 61 lifting of cdga homotopies, 63 localization map, 291 loop product, 342, 343 dual of, 343 Hochschild model, 345 model of, 345 lower grading, 188 Lusternik–Fet theorem, 209 Lusternik–Schnirelmann category, 88

Malcev basis, 121 manifold almost complex, 150 Calabi–Eckmann, 159, 160 closed, 227 cohomologically symplectic, 188 complex, 148 elliptic four-dimensional, 108 flat, 241 Grassmann, 177 Heisenberg, 117, 162 Hodge, 160 Hopf, 159, 161, 174 integrable almost complex, 154 Kähler, 157, 166 strict formality of, 172 Kodaira–Thurston, 120, 156, 162, 165, 190 orientable, 397 P-, 232 pure, 188 symplectic, 182 nonformal, 338 mapping cone, 429 algebraic, 322 cylinder, 429 space, 370 pointed, 373 theorem, 89 Massey product, 94 higher order, 95 maximal torus, 8 and Weyl group, 127 McDuff’s blow-up example, 337 minimal cdga, 59 realization of, 90 minimal model, 59, 64 of free loop space of S2 , 213 Q, 67 Q to R, 68 R, 67 canonical, 140, 141 equivariant, 123, 124, 221 existence, 64 of H-spaces, 73 of a product of manifolds, 73 of biquotient purity of, 138 of classifying space, 72 of cofiber, 90

Index

of complex algebraic variety, 367 of complex projective space, 72, 78 of connected sum, 108 of diagonal map, 73 of fiber, 81 of homogeneous space, 131, 133 of Lie group, 71 of map, 66, 67 of nilmanifold, 118, 122 of space, 67 of sphere, 72, 78 of Stiefel manifold, 71 of torus, 72 of wedge of spaces, 73 rational, 67 of map, 67 real, 67 relative, 66, 67 uniqueness, 64 of Kodaira–Thurston manifold, 162 mixed Hodge structure, 366 model, 74 r-filtered, 171 bigraded, 95, 113 Cartan–Weil, 51 Cartan–Weil, 309 Dolbeault, 169 filtered, 96 finite, 75 for blow-up, 336, 337 for configuration space, 355 for loop product, 345 Haefliger, 369 for mapping space, 369 of Map(CP2 , S6 ), 373 of biquotient, 137 of blow-up example, 334 of complement of submanifold, 324 of free loop space, 210, 369 of homogeneous space, 83, 131, 133 of map, 74 of McDuff blow-up example, 337 of principal bundle, 84 of space, 74 Poincaré duality, 109 pure, 86, 188 r-bigraded, 170 semifree, 101 Sullivan, 64 moment-angle complex, 363 morphism, 58

NDR-pair, 429 Nijenhuis tensor, 155 nilmanifold, 92, 116 almost complex structure on, 155 four-dimensional, 192 Lefschetz type, 193 minimal model of, 118, 122 nonformality of, 120, 121 non-Kähler, 167 symplectic, 191 and Brylinski conjecture, 185 toral rank of, 288 two-step, 290 nilpotent fibration, 79 fundamental group, 76 group, 76 space, 68 symplectic, 194 non-degenerate basepoint, 429 non-formal space, 94 normalization of the bar construction, 382 normalizer, 7 obstructions to formality, 97 one-parameter subgroup, 5 orbit, 274 orientable, 397 orthogonal group, 9 P-manifold, 232 rationally elliptic, 234 parallelizable, 5 path fibration, 39, 79, 431 of S3 , 414 Poincaré duality isomorphism, 319 model, 109 polar decomposition, 3 Pontryagin class, 46 Postnikov tower, 88 for nilmanifold, 122 primitive element, 18 principal bundle, 32 flat, 254 pullback of, 33 trivial, 33 universal, 34, 275 problem Belegradek–Kapovitch biquotients, 390

457

458

Index

derivations, 390 curvature and cohomology, 390 elliptic fiber, 389 exponential growth of Betti numbers, 391 geometry versus algebra, 390 Lalonde–McDuff and models, 391 pullback of fiber bundle, 27 of principal bundle, 33 Puppe sequence, 40, 432 pure cdga, 419 fibration, 189 manifold, 188 minimal cdga, 86, 188 splitting of, 87 pure Hodge structure, 364 and bigradation of model, 365 quasi-isomorphism, 59, 60 quasi-nilpotent fibration, 79 question Lalonde–McDuff, 314, 391 r-bigraded cdga, 170 rank, 8, 121 rational category, 89 homotopy descent phenomena, 249 Lie algebra, 77 of fixed point set, 301 type, 68 space, 90 toral rank, 272 of nilmanifold, 288 rationalization of space, 68, 91 rationally elliptic space, 84 hyperbolic space, 84 real homotopy type, 69, 143 realization of graded Lie algebra, 92 of minimal cdga, 90 of model by manifold, 106 relative atomic complex, 359 relative minimal cdga, 66 relative minimal model, 67 of diagonal map, 73 of fibration, 81

of Hopf fibration, 82 of path fibration, 81 of sphere bundle, 82 of universal principal bundle, 82 r-bigraded, 170 realization of, 91 relative model of Borel fibration, 280 Ricci curvature, 269 Riemann curvature, 240 Riemannian connection, 239 properties of, 239 metric, 239 scalar curvature, 269 sectional curvature, 240 semifree, 236 model, 101, 320 module, 100 shriek map, 319, 329, 342 cohomological, 319 shuffle, 381 product, 384 singular cohomology, 403 soul, 252 space mapping, 370 non-nilpotent, 68 of orbits, 274 symmetric, 53 of free paths, 377 special orthogonal group, 9 special unitary group, 10 spectral sequence, 409 associated to double complex, 421 collapses, 410 comparison theorem, 416 converges, 410 first quadrant, 410 in cohomology, 411 morphism, 409, 411 odd, 419 of Hopf fibration, 414 Serre, 413 transgression, 411 spinor group, 9 splitting rigid, 259 and derivations, 263 Stiefel manifold, 42 cohomology of, 43, 45 minimal model of, 71 Stokes’s theorem, 402

Index

strict formality, 172 subspace arrangement formality of, 361 Sullivan model, 64 Sullivan cdga, 59, 68 minimal, 59 structure of, 60 Sullivan polynomial forms, 67 suspension of graded vector space, 57 symmetric space, 53 symplectic action, 306 splitting of, 310 bundle, 190 group, 11 manifold, 182 almost complex structure of, 187 nonformal, 338 nilpotent manifold, 194 star operator, 184 vector space, 200 symplectically harmonic form, 184 tangent bundle, 393, 396 map, 393 space complex, 153 real, 152 theorem Allday–Halperin, 279 Babenko–Taimanov, 338 Belegradek–Kapovitch, 259 Blanchard, 165 Borel localization, 272, 292 Chen, 388 Cohen–Jones, 347 Darboux, 182 de Rham, 403 Eschenburg, 243 existence of equivariant model, 124 Fang–Rong, 244 Gromoll–Meyer, 210 Gromov, 224 Grove, 216 Haefliger, 369 Hodge decomposition, 407 Hsiang, 276 Kriz, Totaro, 355 Lambrechts, 225

Lusternik–Fet, 209 mapping, 89 Miller, 110 second proof of, 115 Nomizu, 119, 191 O’Neill, 242 of the cube, 428 on filtered algebras and spectral sequences, 412 Paternain, 230 soul, 252 splitting, 252 Stasheff, 113 Stepien, 315 Stokes’s, 402 Sullivan–Barge, 106 Tanaka, 222 Thurston, 190 Totaro, 250 Vigué-Poirrier–Sullivan, 213 Whitehead, 424 Yuzvinsky, 360 Zeeman–Moore, 416 hard Lefschetz, 163 TNCZ, 164, 165, 293, 313 criterion for, 293 topological blow-up, 332 topological entropy, 227 and rational ellipticity, 230 toral rank, 271, 277 of wedge of spheres, 277 rational, 272, 278 of homogeneous space, 279 of Lie group, 279 toric variety, 363 total complex, 421 total degree, 169, 409 totally convex, 252 geodesic, 252 non-cohomologous to zero, 164, 293 transgressive element, 39 TRC, 283, 388 algebraic, 284 and hard Lefschetz, 285 and Lefschetz type, 312 for homogeneous spaces, 284 for Lie algebras, 289 is toral rank conjecture, 272 unitary group, 10 universal fibration, 45

459